Bowen Liang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Bowen Liang has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
- Huabo Zhu
School of Mechanical Engineering, Hebei University of Technology, Tianjin, China
4 shared publications - Yourui Tao
State Key Laboratory of Reliability and Intelligence of Electrical Equipment, Hebei University of Technology, Tianjin, China
4 shared publications - Shijie Guo
College of Intelligent Robotics and Advanced Manufacturing, Fudan University, Shanghai, China
2 shared publications - Song Yao
Hebei Key Laboratory of Robot Perception and Human-robot Interaction, Hebei University of Technology, Tianjin, China
2 shared publications Xu Han
State Key Laboratory of Reliability and Intelligence of Electrical Equipment Jointly Constructed by the Ministry and the Province, Hebei University of Technology, Tianjin, People’s Republic of China
2 shared publicationsYao Song
Hebei Univ Technol, Hebei Key Lab Robot Percept & Human Robot Interact, Tianjin, Peoples R China
2 shared publications
Latest Publications
- 17 July 2025
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 36 (7) , 075022
Generalized differentiable Perspective-n-Point without 2D-3D correspondences for pose measurement
Solving the Perspective-n-Points (PnP) problem without two-dimensional (2D)-three-dimensional (3D) correspondences is known as blind PnP. The extensive search space and numerous outliers present significant challenges in accurately measuring the camera pose. We proposed a generalized differentiable blind PnP method for real-time, high-precision camera pose estimation and measurement. The method introduces an innovative graph neural network for cross-modal matching of 2D and 3D points, jointly identifying correspondences and rejecting outliers. Assignments are estimated by solving a differentiable optimal transport problem, with dustbin channels added to handle outliers effectively. The probabilistic PnP layer facilitates end-to-end learning and inference of pose, leveraging 2D-3D joint correspondence probabilities. Furthermore, geometric constraints and weighted nonlinear PnP solvers are employed to optimize the assignment matrix globally. Experimental results show that our method surpasses existing algorithms in accuracy and speed for blind PnP problems, particularly with large-scale datasets, highlighting its potential as a generalized PnP solver. In comparison to laser trackers for robot arm positioning, the proposed method exhibits an error level of 10−1 mm, suggesting its viability as a cost-effective and efficient substitute.
- 17 July 2025
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 36 (7) , 075022
Generalized differentiable Perspective-n-Point without 2D-3D correspondences for pose measurement
Solving the Perspective-n-Points (PnP) problem without two-dimensional (2D)-three-dimensional (3D) correspondences is known as blind PnP. The extensive search space and numerous outliers present significant challenges in accurately measuring the camera pose. We proposed a generalized differentiable blind PnP method for real-time, high-precision camera pose estimation and measurement. The method introduces an innovative graph neural network for cross-modal matching of 2D and 3D points, jointly identifying correspondences and rejecting outliers. Assignments are estimated by solving a differentiable optimal transport problem, with dustbin channels added to handle outliers effectively. The probabilistic PnP layer facilitates end-to-end learning and inference of pose, leveraging 2D-3D joint correspondence probabilities. Furthermore, geometric constraints and weighted nonlinear PnP solvers are employed to optimize the assignment matrix globally. Experimental results show that our method surpasses existing algorithms in accuracy and speed for blind PnP problems, particularly with large-scale datasets, highlighting its potential as a generalized PnP solver. In comparison to laser trackers for robot arm positioning, the proposed method exhibits an error level of 10−1 mm, suggesting its viability as a cost-effective and efficient substitute.
- 13 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (12) , 22488-22500
Robust Odometry for Wheeled Robots in Smoky Environments: Leveraging Smoke-Adaptive Image Features and Multisensor Tight Coupling
Odometry provides state estimation for the robot, which is crucial for enabling autonomous mobility in firefighting robots. However, in visually degraded environments such as haze and smoke, conventional odometry is prone to failure. In this study, we propose a visual-inertial-Wheel encoders odometry system, VIW-Haze, specifically designed for firefighting robots. This system integrates proprioceptive sensors (encoders, inertial measurement units) with visual odometry. To address the interference caused by smoke and haze, we designed a feature extraction network with a smoke-adaptive module. Additionally, we proposed a method to tightly integrate encoder measurements by modeling the slip rate as a scale factor. We incorporated slip-adjusted encoder measurements during initialization, reducing the system’s dependence on motion excitation and parallax. Finally, we conducted extensive testing on various datasets to validate our approach. The results indicate that the feature extraction network can effectively detect feature points in smoke images, demonstrating excellent keypoint repeatability. Additionally, due to the real-time estimation of slip, our odometry outperforms traditional methods, with accuracy improvements exceeding 20%. It also remains stable even when visual input fails entirely. Importantly, this approach is not limited to firefighting robots; it can provide reliable state estimation for any ground robot operating in visually degraded environments.
- 13 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (12) , 22488-22500
Robust Odometry for Wheeled Robots in Smoky Environments: Leveraging Smoke-Adaptive Image Features and Multisensor Tight Coupling
Odometry provides state estimation for the robot, which is crucial for enabling autonomous mobility in firefighting robots. However, in visually degraded environments such as haze and smoke, conventional odometry is prone to failure. In this study, we propose a visual-inertial-Wheel encoders odometry system, VIW-Haze, specifically designed for firefighting robots. This system integrates proprioceptive sensors (encoders, inertial measurement units) with visual odometry. To address the interference caused by smoke and haze, we designed a feature extraction network with a smoke-adaptive module. Additionally, we proposed a method to tightly integrate encoder measurements by modeling the slip rate as a scale factor. We incorporated slip-adjusted encoder measurements during initialization, reducing the system’s dependence on motion excitation and parallax. Finally, we conducted extensive testing on various datasets to validate our approach. The results indicate that the feature extraction network can effectively detect feature points in smoke images, demonstrating excellent keypoint repeatability. Additionally, due to the real-time estimation of slip, our odometry outperforms traditional methods, with accuracy improvements exceeding 20%. It also remains stable even when visual input fails entirely. Importantly, this approach is not limited to firefighting robots; it can provide reliable state estimation for any ground robot operating in visually degraded environments.
- 26 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (7) , 11226-11235
Noninvasive Respiratory Monitoring Through Correlation of Thoracoabdominal Movements and Pressure Fluctuations
This study presents a novel, non-invasive system that quantifies thoracoabdominal respiratory movements and correlates them with back pressure fluctuations, advancing respiratory monitoring. A key contribution is the development of a method to measure the three-dimensional movements of the chest and abdomen using a 3D motion capture system, enabling high-precision analysis of respiratory dynamics. Additionally, another significant contribution of this work is the development of a novel correlation model that integrates the thoracoabdominal motion data with back pressure fluctuations recorded by a tactile sensor array. This integration represents a critical departure from previous research, which often examines these parameters in isolation. By linking back pressure fluctuations to respiratory displacements, our method facilitates non-contact, unobtrusive respiratory monitoring, offering a practical and patient-friendly alternative to traditional contact-based techniques such as Respiratory Inductance Plethysmography (RIP). Furthermore, this study identifies optimal regions for pressure monitoring, which enhances the system’s precision and usability for long-term monitoring in clinical and home environments. This step addresses a critical gap in the current literature, where practical implementation strategies are often overlooked. The results reveal a clear correlation between pressure fluctuations and respiratory movements, forming the basis for a model that can predict respiratory states. This system provides significant potential for practical application, offering a promising solution for reliable, unobtrusive respiratory monitoring in both clinical and home settings.
- 26 February 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (7) , 11226-11235
Noninvasive Respiratory Monitoring Through Correlation of Thoracoabdominal Movements and Pressure Fluctuations
This study presents a novel, non-invasive system that quantifies thoracoabdominal respiratory movements and correlates them with back pressure fluctuations, advancing respiratory monitoring. A key contribution is the development of a method to measure the three-dimensional movements of the chest and abdomen using a 3D motion capture system, enabling high-precision analysis of respiratory dynamics. Additionally, another significant contribution of this work is the development of a novel correlation model that integrates the thoracoabdominal motion data with back pressure fluctuations recorded by a tactile sensor array. This integration represents a critical departure from previous research, which often examines these parameters in isolation. By linking back pressure fluctuations to respiratory displacements, our method facilitates non-contact, unobtrusive respiratory monitoring, offering a practical and patient-friendly alternative to traditional contact-based techniques such as Respiratory Inductance Plethysmography (RIP). Furthermore, this study identifies optimal regions for pressure monitoring, which enhances the system’s precision and usability for long-term monitoring in clinical and home environments. This step addresses a critical gap in the current literature, where practical implementation strategies are often overlooked. The results reveal a clear correlation between pressure fluctuations and respiratory movements, forming the basis for a model that can predict respiratory states. This system provides significant potential for practical application, offering a promising solution for reliable, unobtrusive respiratory monitoring in both clinical and home settings.
- 4 November 2024
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 36 (1) , 015702
DeepLPos: a comprehensive hybrid deep learning model for lying position recognition using a tactile sensor array system
Unpredictable limb movements or turning motions can significantly disrupt the accurate extraction of physiological signals, such as respiratory and heart rates. In clinical environments, reliable detection of lying positions is crucial for continuous patient monitoring, particularly during sleep. In this paper, a smart sleeping position recognition system is proposed, which employs a tactile pressure sensor array based on the unique structure of ‘the electrostatic double-layer capacitors’. The sensor array, comprising 64 rows and 32 columns (2048 nodes), captures four types of healthy lying positions using an 8-bit AD module. Despite challenges arising from limited experimental samples for accurate training, we propose DeepLPos, a hybrid deep learning approach combining generative adversarial networks and the you only look once network. To tackle the differentiation challenge between supine and prone positions, we introduce an SPD Conv attention module to enhance the resolution of detailed descriptions in pressure images. The model is further pruned to optimize both structure and parameters, enabling efficient real-time detection. Evaluated on the SLP dataset, the proposed system achieves an accuracy of 97.5% with a real-time processing speed of 0.069 s per frame, demonstrating its potential for practical, high-precision measurement and monitoring applications in healthcare.
- 4 November 2024
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 36 (1) , 015702
DeepLPos: a comprehensive hybrid deep learning model for lying position recognition using a tactile sensor array system
Unpredictable limb movements or turning motions can significantly disrupt the accurate extraction of physiological signals, such as respiratory and heart rates. In clinical environments, reliable detection of lying positions is crucial for continuous patient monitoring, particularly during sleep. In this paper, a smart sleeping position recognition system is proposed, which employs a tactile pressure sensor array based on the unique structure of ‘the electrostatic double-layer capacitors’. The sensor array, comprising 64 rows and 32 columns (2048 nodes), captures four types of healthy lying positions using an 8-bit AD module. Despite challenges arising from limited experimental samples for accurate training, we propose DeepLPos, a hybrid deep learning approach combining generative adversarial networks and the you only look once network. To tackle the differentiation challenge between supine and prone positions, we introduce an SPD Conv attention module to enhance the resolution of detailed descriptions in pressure images. The model is further pruned to optimize both structure and parameters, enabling efficient real-time detection. Evaluated on the SLP dataset, the proposed system achieves an accuracy of 97.5% with a real-time processing speed of 0.069 s per frame, demonstrating its potential for practical, high-precision measurement and monitoring applications in healthcare.
- 30 January 2024
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 35 (4) , 046308
Advancing robust state estimation of wheeled robots in degenerate environments: harnessing ground manifold and motion states
State estimation is crucial for enabling autonomous mobility in mobile robots. However, traditional localization methods often falter in degraded environments, including issues like visual occlusion, lidar performance degradation, and global navigation satellite system signal interference. This paper presents a novel estimation approach for wheeled robots, exclusively utilizing proprioceptive sensors such as encoders and inertial measurement units (IMU). Initially, the motion manifolds extracted from the historical trajectories are used to assist the encoder in realizing the orientation estimation. Furthermore, a hybrid neural network is designed to categorize the robot’s operational state, and the corresponding pseudo-constraints are added to improve the estimation accuracy. We utilize an error state Kalman filter for the encoder and IMU data fusion. Lastly, comprehensive testing is conducted using both datasets and real-world robotic platforms. The findings underscore that the integration of manifold and motion constraints within our proposed state estimator substantially elevates accuracy compared to conventional approaches. Compare with the methods commonly used in engineering, the accuracy of this method is improved by more than 20%. Crucially, this methodology enables dependable estimation even in degraded environments.
- 30 January 2024
- Published by IOP Publishing in Measurement Science and Technology
- Vol. 35 (4) , 046308
Advancing robust state estimation of wheeled robots in degenerate environments: harnessing ground manifold and motion states
State estimation is crucial for enabling autonomous mobility in mobile robots. However, traditional localization methods often falter in degraded environments, including issues like visual occlusion, lidar performance degradation, and global navigation satellite system signal interference. This paper presents a novel estimation approach for wheeled robots, exclusively utilizing proprioceptive sensors such as encoders and inertial measurement units (IMU). Initially, the motion manifolds extracted from the historical trajectories are used to assist the encoder in realizing the orientation estimation. Furthermore, a hybrid neural network is designed to categorize the robot’s operational state, and the corresponding pseudo-constraints are added to improve the estimation accuracy. We utilize an error state Kalman filter for the encoder and IMU data fusion. Lastly, comprehensive testing is conducted using both datasets and real-world robotic platforms. The findings underscore that the integration of manifold and motion constraints within our proposed state estimator substantially elevates accuracy compared to conventional approaches. Compare with the methods commonly used in engineering, the accuracy of this method is improved by more than 20%. Crucially, this methodology enables dependable estimation even in degraded environments.
- 6 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 31 (10709908) , 21-25
Sub-Pixel Checkerboard Corner Localization for Robust Vision Measurement
Precise localization of checkerboard corner is crucial for camera calibration and accurate vision measurement. The accuracy of classical corner localization methods is affected by environmental factors, which hinders the effectiveness of vision measurements in complex scenes. We propose a novel end-to-end sub-pixel method for localizing checkerboard corners to overcome this challenge, demonstrating improved robustness in adverse conditions. We employ an EfficientNetv2-like backbone network to map the input image to a set of corners using offset regression. The continuous heatmap matching loss enables fine-grained spatial learning ability, while score loss ensures prediction sparsity and suppresses false positive probabilities. The model is trained on synthesized datasets with precise corner labels. Quantitative experimental results demonstrate the superior sub-pixel localization precision and robustness of our method on both synthetic and real-world datasets. The code is publicly available at https://github.com/huabo-zhu/LSCCL .
- 6 December 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 31 (10709908) , 21-25
Sub-Pixel Checkerboard Corner Localization for Robust Vision Measurement
Precise localization of checkerboard corner is crucial for camera calibration and accurate vision measurement. The accuracy of classical corner localization methods is affected by environmental factors, which hinders the effectiveness of vision measurements in complex scenes. We propose a novel end-to-end sub-pixel method for localizing checkerboard corners to overcome this challenge, demonstrating improved robustness in adverse conditions. We employ an EfficientNetv2-like backbone network to map the input image to a set of corners using offset regression. The continuous heatmap matching loss enables fine-grained spatial learning ability, while score loss ensures prediction sparsity and suppresses false positive probabilities. The model is trained on synthesized datasets with precise corner labels. Quantitative experimental results demonstrate the superior sub-pixel localization precision and robustness of our method on both synthetic and real-world datasets. The code is publicly available at https://github.com/huabo-zhu/LSCCL .