font
Filter
×
2026
Hafez, Muhammad Burhan; Miller, Emily; Milford, Michael J.; Ramchurn, Gopal; Ehsan, Shoaib
Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 11, no. 5, pp. 5899–5906, 2026.
Abstract | Links | BibTeX | Tags: Deep Learning for Visual Perception, Localization, Vision-Based Navigation
@article{soton510726,
title = {Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition},
author = {Muhammad Burhan Hafez and Emily Miller and Michael J. Milford and Gopal Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/510726/},
year = {2026},
date = {2026-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {11},
number = {5},
pages = {5899–5906},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>},
keywords = {Deep Learning for Visual Perception, Localization, Vision-Based Navigation},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>
2025
Ismagilov, Timur; Ferrarini, Bruno; Milford, Michael; Tuyen, Nguyen Tan Viet; Ramchurn, Sarvapali D.; Ehsan, Shoaib
On motion blur and deblurring in visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 5, pp. 4746–4753, 2025.
Abstract | Links | BibTeX | Tags: data sets for robotic vision, Localization, Vision-Based Navigation
@article{soton507348,
title = {On motion blur and deblurring in visual place recognition},
author = {Timur Ismagilov and Bruno Ferrarini and Michael Milford and Nguyen Tan Viet Tuyen and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/507348/},
year = {2025},
date = {2025-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {5},
pages = {4746–4753},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>},
keywords = {data sets for robotic vision, Localization, Vision-Based Navigation},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>
Grainge, Oliver Edward; Milford, Michael J.; Bodala, Indu; Ramchurn, Sarvapali D.; Ehsan, Shoaib
Structured pruning for efficient visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 2, pp. 2024–2031, 2025.
Abstract | Links | BibTeX | Tags: accuracy, computational modelling, convolutional neural network, feature extraction, Localization, memory management, Real-time systems, representation learning, robustness, Vision-Based Navigation, visual place recognition (VPR), visualization
@article{soton502843,
title = {Structured pruning for efficient visual place recognition},
author = {Oliver Edward Grainge and Michael J. Milford and Indu Bodala and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/502843/},
year = {2025},
date = {2025-02-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {2},
pages = {2024–2031},
abstract = {Visual Place Recognition (VPR) is fundamental for the global re-localization of robots and devices, enabling them to recognize previously visited locations based on visual inputs. This capability is crucial for maintaining accurate mapping and localization over large areas. Given that VPR methods need to operate in real-time on embedded systems, it is critical to optimize these systems for minimal resource consumption. While the most efficient VPR approaches employ standard convolutional backbones with fixed descriptor dimensions, these often lead to redundancy in the embedding space as well as in the network architecture. Our work introduces a novel structured pruning method, to not only streamline common VPR architectures but also to strategically remove redundancies within the feature embedding space. This dual focus significantly enhances the efficiency of the system, reducing both map and model memory requirements and decreasing feature extraction and retrieval latencies. Our approach has reduced memory usage and latency by 21% and 16%, respectively, across models, while minimally impacting recall@1 accuracy by less than 1%. This significant improvement enhances real-time applications on edge devices with negligible accuracy loss.},
keywords = {accuracy, computational modelling, convolutional neural network, feature extraction, Localization, memory management, Real-time systems, representation learning, robustness, Vision-Based Navigation, visual place recognition (VPR), visualization},
pubstate = {published},
tppubtype = {article}
}
Visual Place Recognition (VPR) is fundamental for the global re-localization of robots and devices, enabling them to recognize previously visited locations based on visual inputs. This capability is crucial for maintaining accurate mapping and localization over large areas. Given that VPR methods need to operate in real-time on embedded systems, it is critical to optimize these systems for minimal resource consumption. While the most efficient VPR approaches employ standard convolutional backbones with fixed descriptor dimensions, these often lead to redundancy in the embedding space as well as in the network architecture. Our work introduces a novel structured pruning method, to not only streamline common VPR architectures but also to strategically remove redundancies within the feature embedding space. This dual focus significantly enhances the efficiency of the system, reducing both map and model memory requirements and decreasing feature extraction and retrieval latencies. Our approach has reduced memory usage and latency by 21% and 16%, respectively, across models, while minimally impacting recall@1 accuracy by less than 1%. This significant improvement enhances real-time applications on edge devices with negligible accuracy loss.
1.
Hafez, Muhammad Burhan; Miller, Emily; Milford, Michael J.; Ramchurn, Gopal; Ehsan, Shoaib
Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 11, no. 5, pp. 5899–5906, 2026.
@article{soton510726,
title = {Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition},
author = {Muhammad Burhan Hafez and Emily Miller and Michael J. Milford and Gopal Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/510726/},
year = {2026},
date = {2026-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {11},
number = {5},
pages = {5899–5906},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>
2.
Ismagilov, Timur; Ferrarini, Bruno; Milford, Michael; Tuyen, Nguyen Tan Viet; Ramchurn, Sarvapali D.; Ehsan, Shoaib
On motion blur and deblurring in visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 5, pp. 4746–4753, 2025.
@article{soton507348,
title = {On motion blur and deblurring in visual place recognition},
author = {Timur Ismagilov and Bruno Ferrarini and Michael Milford and Nguyen Tan Viet Tuyen and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/507348/},
year = {2025},
date = {2025-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {5},
pages = {4746–4753},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>
3.
Grainge, Oliver Edward; Milford, Michael J.; Bodala, Indu; Ramchurn, Sarvapali D.; Ehsan, Shoaib
Structured pruning for efficient visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 2, pp. 2024–2031, 2025.
@article{soton502843,
title = {Structured pruning for efficient visual place recognition},
author = {Oliver Edward Grainge and Michael J. Milford and Indu Bodala and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/502843/},
year = {2025},
date = {2025-02-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {2},
pages = {2024–2031},
abstract = {Visual Place Recognition (VPR) is fundamental for the global re-localization of robots and devices, enabling them to recognize previously visited locations based on visual inputs. This capability is crucial for maintaining accurate mapping and localization over large areas. Given that VPR methods need to operate in real-time on embedded systems, it is critical to optimize these systems for minimal resource consumption. While the most efficient VPR approaches employ standard convolutional backbones with fixed descriptor dimensions, these often lead to redundancy in the embedding space as well as in the network architecture. Our work introduces a novel structured pruning method, to not only streamline common VPR architectures but also to strategically remove redundancies within the feature embedding space. This dual focus significantly enhances the efficiency of the system, reducing both map and model memory requirements and decreasing feature extraction and retrieval latencies. Our approach has reduced memory usage and latency by 21% and 16%, respectively, across models, while minimally impacting recall@1 accuracy by less than 1%. This significant improvement enhances real-time applications on edge devices with negligible accuracy loss.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Visual Place Recognition (VPR) is fundamental for the global re-localization of robots and devices, enabling them to recognize previously visited locations based on visual inputs. This capability is crucial for maintaining accurate mapping and localization over large areas. Given that VPR methods need to operate in real-time on embedded systems, it is critical to optimize these systems for minimal resource consumption. While the most efficient VPR approaches employ standard convolutional backbones with fixed descriptor dimensions, these often lead to redundancy in the embedding space as well as in the network architecture. Our work introduces a novel structured pruning method, to not only streamline common VPR architectures but also to strategically remove redundancies within the feature embedding space. This dual focus significantly enhances the efficiency of the system, reducing both map and model memory requirements and decreasing feature extraction and retrieval latencies. Our approach has reduced memory usage and latency by 21% and 16%, respectively, across models, while minimally impacting recall@1 accuracy by less than 1%. This significant improvement enhances real-time applications on edge devices with negligible accuracy loss.
2026
Hafez, Muhammad Burhan; Miller, Emily; Milford, Michael J.; Ramchurn, Gopal; Ehsan, Shoaib
Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 11, no. 5, pp. 5899–5906, 2026.
Abstract | Links | BibTeX | Tags: Deep Learning for Visual Perception, Localization, Vision-Based Navigation
@article{soton510726,
title = {Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition},
author = {Muhammad Burhan Hafez and Emily Miller and Michael J. Milford and Gopal Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/510726/},
year = {2026},
date = {2026-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {11},
number = {5},
pages = {5899–5906},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>},
keywords = {Deep Learning for Visual Perception, Localization, Vision-Based Navigation},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>
2025
Ismagilov, Timur; Ferrarini, Bruno; Milford, Michael; Tuyen, Nguyen Tan Viet; Ramchurn, Sarvapali D.; Ehsan, Shoaib
On motion blur and deblurring in visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 5, pp. 4746–4753, 2025.
Abstract | Links | BibTeX | Tags: data sets for robotic vision, Localization, Vision-Based Navigation
@article{soton507348,
title = {On motion blur and deblurring in visual place recognition},
author = {Timur Ismagilov and Bruno Ferrarini and Michael Milford and Nguyen Tan Viet Tuyen and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/507348/},
year = {2025},
date = {2025-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {5},
pages = {4746–4753},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>},
keywords = {data sets for robotic vision, Localization, Vision-Based Navigation},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>
Grainge, Oliver Edward; Milford, Michael J.; Bodala, Indu; Ramchurn, Sarvapali D.; Ehsan, Shoaib
Structured pruning for efficient visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 2, pp. 2024–2031, 2025.
Abstract | Links | BibTeX | Tags: accuracy, computational modelling, convolutional neural network, feature extraction, Localization, memory management, Real-time systems, representation learning, robustness, Vision-Based Navigation, visual place recognition (VPR), visualization
@article{soton502843,
title = {Structured pruning for efficient visual place recognition},
author = {Oliver Edward Grainge and Michael J. Milford and Indu Bodala and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/502843/},
year = {2025},
date = {2025-02-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {2},
pages = {2024–2031},
abstract = {Visual Place Recognition (VPR) is fundamental for the global re-localization of robots and devices, enabling them to recognize previously visited locations based on visual inputs. This capability is crucial for maintaining accurate mapping and localization over large areas. Given that VPR methods need to operate in real-time on embedded systems, it is critical to optimize these systems for minimal resource consumption. While the most efficient VPR approaches employ standard convolutional backbones with fixed descriptor dimensions, these often lead to redundancy in the embedding space as well as in the network architecture. Our work introduces a novel structured pruning method, to not only streamline common VPR architectures but also to strategically remove redundancies within the feature embedding space. This dual focus significantly enhances the efficiency of the system, reducing both map and model memory requirements and decreasing feature extraction and retrieval latencies. Our approach has reduced memory usage and latency by 21% and 16%, respectively, across models, while minimally impacting recall@1 accuracy by less than 1%. This significant improvement enhances real-time applications on edge devices with negligible accuracy loss.},
keywords = {accuracy, computational modelling, convolutional neural network, feature extraction, Localization, memory management, Real-time systems, representation learning, robustness, Vision-Based Navigation, visual place recognition (VPR), visualization},
pubstate = {published},
tppubtype = {article}
}
Visual Place Recognition (VPR) is fundamental for the global re-localization of robots and devices, enabling them to recognize previously visited locations based on visual inputs. This capability is crucial for maintaining accurate mapping and localization over large areas. Given that VPR methods need to operate in real-time on embedded systems, it is critical to optimize these systems for minimal resource consumption. While the most efficient VPR approaches employ standard convolutional backbones with fixed descriptor dimensions, these often lead to redundancy in the embedding space as well as in the network architecture. Our work introduces a novel structured pruning method, to not only streamline common VPR architectures but also to strategically remove redundancies within the feature embedding space. This dual focus significantly enhances the efficiency of the system, reducing both map and model memory requirements and decreasing feature extraction and retrieval latencies. Our approach has reduced memory usage and latency by 21% and 16%, respectively, across models, while minimally impacting recall@1 accuracy by less than 1%. This significant improvement enhances real-time applications on edge devices with negligible accuracy loss.
1.
Hafez, Muhammad Burhan; Miller, Emily; Milford, Michael J.; Ramchurn, Gopal; Ehsan, Shoaib
Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 11, no. 5, pp. 5899–5906, 2026.
@article{soton510726,
title = {Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition},
author = {Muhammad Burhan Hafez and Emily Miller and Michael J. Milford and Gopal Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/510726/},
year = {2026},
date = {2026-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {11},
number = {5},
pages = {5899–5906},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>
2.
Ismagilov, Timur; Ferrarini, Bruno; Milford, Michael; Tuyen, Nguyen Tan Viet; Ramchurn, Sarvapali D.; Ehsan, Shoaib
On motion blur and deblurring in visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 5, pp. 4746–4753, 2025.
@article{soton507348,
title = {On motion blur and deblurring in visual place recognition},
author = {Timur Ismagilov and Bruno Ferrarini and Michael Milford and Nguyen Tan Viet Tuyen and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/507348/},
year = {2025},
date = {2025-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {5},
pages = {4746–4753},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>
3.
Grainge, Oliver Edward; Milford, Michael J.; Bodala, Indu; Ramchurn, Sarvapali D.; Ehsan, Shoaib
Structured pruning for efficient visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 2, pp. 2024–2031, 2025.
@article{soton502843,
title = {Structured pruning for efficient visual place recognition},
author = {Oliver Edward Grainge and Michael J. Milford and Indu Bodala and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/502843/},
year = {2025},
date = {2025-02-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {2},
pages = {2024–2031},
abstract = {Visual Place Recognition (VPR) is fundamental for the global re-localization of robots and devices, enabling them to recognize previously visited locations based on visual inputs. This capability is crucial for maintaining accurate mapping and localization over large areas. Given that VPR methods need to operate in real-time on embedded systems, it is critical to optimize these systems for minimal resource consumption. While the most efficient VPR approaches employ standard convolutional backbones with fixed descriptor dimensions, these often lead to redundancy in the embedding space as well as in the network architecture. Our work introduces a novel structured pruning method, to not only streamline common VPR architectures but also to strategically remove redundancies within the feature embedding space. This dual focus significantly enhances the efficiency of the system, reducing both map and model memory requirements and decreasing feature extraction and retrieval latencies. Our approach has reduced memory usage and latency by 21% and 16%, respectively, across models, while minimally impacting recall@1 accuracy by less than 1%. This significant improvement enhances real-time applications on edge devices with negligible accuracy loss.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Visual Place Recognition (VPR) is fundamental for the global re-localization of robots and devices, enabling them to recognize previously visited locations based on visual inputs. This capability is crucial for maintaining accurate mapping and localization over large areas. Given that VPR methods need to operate in real-time on embedded systems, it is critical to optimize these systems for minimal resource consumption. While the most efficient VPR approaches employ standard convolutional backbones with fixed descriptor dimensions, these often lead to redundancy in the embedding space as well as in the network architecture. Our work introduces a novel structured pruning method, to not only streamline common VPR architectures but also to strategically remove redundancies within the feature embedding space. This dual focus significantly enhances the efficiency of the system, reducing both map and model memory requirements and decreasing feature extraction and retrieval latencies. Our approach has reduced memory usage and latency by 21% and 16%, respectively, across models, while minimally impacting recall@1 accuracy by less than 1%. This significant improvement enhances real-time applications on edge devices with negligible accuracy loss.
Selected Publications
Multi-agent signal-less intersection management with dynamic platoon formation
No data available.
https://eprints.soton.ac.uk/478647/
@inproceedings{soton478647,
title = {Multi-agent signal-less intersection management with dynamic platoon formation},
author = {Phuriwat Worrawichaipat and Enrico Gerding and Ioannis Kaparias and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/478647/},
year = {2023},
date = {2023-05-01},
booktitle = {22nd International Conference on Autonomous Agents and Multiagent Systems (29/05/23 - 02/06/23)},
pages = {1542--1550},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
title = {Multi-agent signal-less intersection management with dynamic platoon formation},
author = {Phuriwat Worrawichaipat and Enrico Gerding and Ioannis Kaparias and Sarvapali Ramchurn},
url = {https://eprints.soton.ac.uk/478647/},
year = {2023},
date = {2023-05-01},
booktitle = {22nd International Conference on Autonomous Agents and Multiagent Systems (29/05/23 - 02/06/23)},
pages = {1542--1550},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
AI Foundation Models: initial review, CMA Consultation, TAS Hub Response
No data available.
https://eprints.soton.ac.uk/477553/
@misc{soton477553,
title = {AI Foundation Models: initial review, CMA Consultation, TAS Hub Response},
author = {Joshua Krook and Derek McAuley and Stuart Anderson and John Downer and Peter Winter and Sarvapali D Ramchurn},
url = {https://eprints.soton.ac.uk/477553/},
year = {2023},
date = {2023-06-01},
publisher = {University of Southampton},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
title = {AI Foundation Models: initial review, CMA Consultation, TAS Hub Response},
author = {Joshua Krook and Derek McAuley and Stuart Anderson and John Downer and Peter Winter and Sarvapali D Ramchurn},
url = {https://eprints.soton.ac.uk/477553/},
year = {2023},
date = {2023-06-01},
publisher = {University of Southampton},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
The effect of data visualisation quality and task density on human-swarm interaction
No data available.
https://eprints.soton.ac.uk/479970/
@inproceedings{soton479970,
title = {The effect of data visualisation quality and task density on human-swarm interaction},
author = {Ayodeji Abioye and Mohammad Naiseh and William Hunt and Jediah R Clark and Sarvapali D Ramchurn and Mohammad Soorati},
url = {https://eprints.soton.ac.uk/479970/},
year = {2023},
date = {2023-06-01},
booktitle = {Proceedings of the 2023 32nd IEEE International Conference on Robot and Human Interactive Communication (RO-MAN)},
publisher = {IEEE},
abstract = {Despite the advantages of having robot swarms, human supervision is required for real-world applications. The performance of the human-swarm system depends on several factors including the data availability for the human operators. In this paper, we study the human factors aspect of the human-swarm interaction and investigate how having access to high-quality data can affect the performance of the human-swarm system - the number of tasks completed and the human trust level in operation. We designed an experiment where a human operator is tasked to operate a swarm to identify casualties in an area within a given time period. One group of operators had the option to request high-quality pictures while the other group had to base their decision on the available low-quality images. We performed a user study with 120 participants and recorded their success rate (directly logged via the simulation platform) as well as their workload and trust level (measured through a questionnaire after completing a human-swarm scenario). The findings from our study indicated that the group granted access to high-quality data exhibited an increased workload and placed greater trust in the swarm, thus confirming our initial hypothesis. However, we also found that the number of accurately identified casualties did not significantly vary between the two groups, suggesting that data quality had no impact on the successful completion of tasks.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
title = {The effect of data visualisation quality and task density on human-swarm interaction},
author = {Ayodeji Abioye and Mohammad Naiseh and William Hunt and Jediah R Clark and Sarvapali D Ramchurn and Mohammad Soorati},
url = {https://eprints.soton.ac.uk/479970/},
year = {2023},
date = {2023-06-01},
booktitle = {Proceedings of the 2023 32nd IEEE International Conference on Robot and Human Interactive Communication (RO-MAN)},
publisher = {IEEE},
abstract = {Despite the advantages of having robot swarms, human supervision is required for real-world applications. The performance of the human-swarm system depends on several factors including the data availability for the human operators. In this paper, we study the human factors aspect of the human-swarm interaction and investigate how having access to high-quality data can affect the performance of the human-swarm system - the number of tasks completed and the human trust level in operation. We designed an experiment where a human operator is tasked to operate a swarm to identify casualties in an area within a given time period. One group of operators had the option to request high-quality pictures while the other group had to base their decision on the available low-quality images. We performed a user study with 120 participants and recorded their success rate (directly logged via the simulation platform) as well as their workload and trust level (measured through a questionnaire after completing a human-swarm scenario). The findings from our study indicated that the group granted access to high-quality data exhibited an increased workload and placed greater trust in the swarm, thus confirming our initial hypothesis. However, we also found that the number of accurately identified casualties did not significantly vary between the two groups, suggesting that data quality had no impact on the successful completion of tasks.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Demonstrating performance benefits of human-swarm teaming
Autonomous swarms of robots can bring robustness, scalability and adaptability to safety-critical tasks such as search and rescue but their application is still very limited. Using semi-autonomous swarms with human control can bring robot swarms to real-world applications. Human operators can define goals for the swarm, monitor their performance and interfere with, or overrule, the decisions and behaviour. We present the "Human And Robot Interactive Swarm'' simulator (HARIS) that allows multi-user interaction with a robot swarm and facilitates qualitative and quantitative user studies through simulation of robot swarms completing tasks, from package delivery to search and rescue, with varying levels of human control. In this demonstration, we showcase the simulator by using it to study the performance gain offered by maintaining a "human-in-the-loop'' over a fully autonomous system as an example. This is illustrated in the context of search and rescue, with an autonomous allocation of resources to those in need.
https://eprints.soton.ac.uk/479903/
@inproceedings{soton479903,
title = {Demonstrating performance benefits of human-swarm teaming},
author = {William Hunt and Jack Ryan and Ayodeji O Abioye and Sarvapali D Ramchurn and Mohammad D Soorati},
url = {https://eprints.soton.ac.uk/479903/},
year = {2023},
date = {2023-05-01},
booktitle = {Proceedings of the 2023 International Conference on Autonomous Agents and Multiagent Systems},
pages = {3062--3064},
publisher = {International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS)},
abstract = {Autonomous swarms of robots can bring robustness, scalability and adaptability to safety-critical tasks such as search and rescue but their application is still very limited. Using semi-autonomous swarms with human control can bring robot swarms to real-world applications. Human operators can define goals for the swarm, monitor their performance and interfere with, or overrule, the decisions and behaviour. We present the "Human And Robot Interactive Swarm'' simulator (HARIS) that allows multi-user interaction with a robot swarm and facilitates qualitative and quantitative user studies through simulation of robot swarms completing tasks, from package delivery to search and rescue, with varying levels of human control. In this demonstration, we showcase the simulator by using it to study the performance gain offered by maintaining a "human-in-the-loop'' over a fully autonomous system as an example. This is illustrated in the context of search and rescue, with an autonomous allocation of resources to those in need.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
title = {Demonstrating performance benefits of human-swarm teaming},
author = {William Hunt and Jack Ryan and Ayodeji O Abioye and Sarvapali D Ramchurn and Mohammad D Soorati},
url = {https://eprints.soton.ac.uk/479903/},
year = {2023},
date = {2023-05-01},
booktitle = {Proceedings of the 2023 International Conference on Autonomous Agents and Multiagent Systems},
pages = {3062--3064},
publisher = {International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS)},
abstract = {Autonomous swarms of robots can bring robustness, scalability and adaptability to safety-critical tasks such as search and rescue but their application is still very limited. Using semi-autonomous swarms with human control can bring robot swarms to real-world applications. Human operators can define goals for the swarm, monitor their performance and interfere with, or overrule, the decisions and behaviour. We present the "Human And Robot Interactive Swarm'' simulator (HARIS) that allows multi-user interaction with a robot swarm and facilitates qualitative and quantitative user studies through simulation of robot swarms completing tasks, from package delivery to search and rescue, with varying levels of human control. In this demonstration, we showcase the simulator by using it to study the performance gain offered by maintaining a "human-in-the-loop'' over a fully autonomous system as an example. This is illustrated in the context of search and rescue, with an autonomous allocation of resources to those in need.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
2026
Hafez, Muhammad Burhan; Miller, Emily; Milford, Michael J.; Ramchurn, Gopal; Ehsan, Shoaib
Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 11, no. 5, pp. 5899–5906, 2026.
@article{soton510726,
title = {Through the lens of doubt: robust and efficient uncertainty estimation for visual place recognition},
author = {Muhammad Burhan Hafez and Emily Miller and Michael J. Milford and Gopal Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/510726/},
year = {2026},
date = {2026-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {11},
number = {5},
pages = {5899–5906},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) enables robots and autonomous vehicles to identify previously visited locations by matching current observations against a database of known places. However, VPR systems face significant challenges when deployed across varying visual environments, lighting conditions, seasonal changes, and viewpoints changes. Failure-critical VPR applications, such as loop closure detection in simultaneous localization and mapping (SLAM) pipelines, require robust estimation of place matching uncertainty. We propose three training-free uncertainty metrics that estimate prediction confidence by analyzing inherent statistical patterns in similarity scores from any existing VPR method. Similarity Distribution (SD) quantifies match distinctiveness by measuring score separation between candidates; Ratio Spread (RS) evaluates competitive ambiguity among top-scoring locations; and Statistical Uncertainty (SU) is a combination of SD and RS that provides a unified metric that generalizes across datasets and VPR methods without requiring validation data to select the optimal metric. All three metrics operate without additional model training, architectural modifications, or computationally expensive geometric verification. Comprehensive evaluation across nine state-of-the-art VPR methods and six benchmark datasets confirms that our metrics excel at discriminating between correct and incorrect VPR matches, and consistently outperform existing approaches while maintaining negligible computational overhead, making it deployable for real-time robotic applications across varied environmental conditions with improved precision-recall performance.ensuremath</pensuremath>
2025
Ismagilov, Timur; Ferrarini, Bruno; Milford, Michael; Tuyen, Nguyen Tan Viet; Ramchurn, Sarvapali D.; Ehsan, Shoaib
On motion blur and deblurring in visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 5, pp. 4746–4753, 2025.
@article{soton507348,
title = {On motion blur and deblurring in visual place recognition},
author = {Timur Ismagilov and Bruno Ferrarini and Michael Milford and Nguyen Tan Viet Tuyen and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/507348/},
year = {2025},
date = {2025-03-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {5},
pages = {4746–4753},
abstract = {ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
ensuremath<pensuremath>Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This letter bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the letter proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.ensuremath</pensuremath>
Grainge, Oliver Edward; Milford, Michael J.; Bodala, Indu; Ramchurn, Sarvapali D.; Ehsan, Shoaib
Structured pruning for efficient visual place recognition Journal Article
In: IEEE Robotics and Automation Letters, vol. 10, no. 2, pp. 2024–2031, 2025.
@article{soton502843,
title = {Structured pruning for efficient visual place recognition},
author = {Oliver Edward Grainge and Michael J. Milford and Indu Bodala and Sarvapali D. Ramchurn and Shoaib Ehsan},
url = {https://eprints.soton.ac.uk/502843/},
year = {2025},
date = {2025-02-01},
journal = {IEEE Robotics and Automation Letters},
volume = {10},
number = {2},
pages = {2024–2031},
abstract = {Visual Place Recognition (VPR) is fundamental for the global re-localization of robots and devices, enabling them to recognize previously visited locations based on visual inputs. This capability is crucial for maintaining accurate mapping and localization over large areas. Given that VPR methods need to operate in real-time on embedded systems, it is critical to optimize these systems for minimal resource consumption. While the most efficient VPR approaches employ standard convolutional backbones with fixed descriptor dimensions, these often lead to redundancy in the embedding space as well as in the network architecture. Our work introduces a novel structured pruning method, to not only streamline common VPR architectures but also to strategically remove redundancies within the feature embedding space. This dual focus significantly enhances the efficiency of the system, reducing both map and model memory requirements and decreasing feature extraction and retrieval latencies. Our approach has reduced memory usage and latency by 21% and 16%, respectively, across models, while minimally impacting recall@1 accuracy by less than 1%. This significant improvement enhances real-time applications on edge devices with negligible accuracy loss.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Visual Place Recognition (VPR) is fundamental for the global re-localization of robots and devices, enabling them to recognize previously visited locations based on visual inputs. This capability is crucial for maintaining accurate mapping and localization over large areas. Given that VPR methods need to operate in real-time on embedded systems, it is critical to optimize these systems for minimal resource consumption. While the most efficient VPR approaches employ standard convolutional backbones with fixed descriptor dimensions, these often lead to redundancy in the embedding space as well as in the network architecture. Our work introduces a novel structured pruning method, to not only streamline common VPR architectures but also to strategically remove redundancies within the feature embedding space. This dual focus significantly enhances the efficiency of the system, reducing both map and model memory requirements and decreasing feature extraction and retrieval latencies. Our approach has reduced memory usage and latency by 21% and 16%, respectively, across models, while minimally impacting recall@1 accuracy by less than 1%. This significant improvement enhances real-time applications on edge devices with negligible accuracy loss.