Jinlei Zhuang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Jinlei Zhuang has a total of 10 co-authors. Here are the top co-authors by number of shared publications:
Weiqiang Ma
Hit Wuhu Robot Technology Research Institute, Wuhu, China
2 shared publications- Ziteng Wang
School of Mechatronics Engineering, Harbin Institute of Technology, Harbin, China
2 shared publications - Anguo Chen
Harbin Institute of Technology,School of Mechatronics Engineering,Harbin,China
1 shared publication Chuqing Cao
Wuhu HIT Robot Technology Research Institute Co., Ltd., Wuhu, China
1 shared publicationKe Wang
State Key Laboratory of Robotics and System, Harbin Institute of Technology, Harbin, China
1 shared publicationRuifeng Li
State Key Laboratory of Robotics and System, Harbin Institute of Technology, Harbin, China
1 shared publication
Latest Publications
- 3 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (24) , 44839-44849
BT-VLoc: Toward Robust and Accurate Visual Localization With Bidirectional Transformer
Visual localization, a fundamental task of numerous computer vision applications, is solved by scene coordinates regression (SCoRe) techniques. These methods leverage neural networks to predict scene coordinates, followed by a PnP algorithm for camera pose recovery. However, they face performance degradation due to ambiguous visual patterns. Recent methods attempt to solve this issue by integrating global and local features, yet the full potential of bidirectional interactions between these features remains underexplored. In this article, we introduce a visual localization framework based on the bidirectional transformer (BFormer), which adaptively models and integrates global and local features through bidirectional interaction. By employing self-modulated convolutions for local feature extraction and self-attention mechanisms in a downsampled space for global information modeling, BFormer enables a dynamic bidirectional adaptation process that strengthens the interplay between local and global representations. As a result, BFormer effectively distinguishes similar image patches by leveraging the complementary strengths of low-level features that provide geometric details and high-level features that contain semantic information. Benefited from BFormer, BTVLoc realizes impressive localization performance on all 7-Scenes, LIVL, and Cambridge datasets.
- 3 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (24) , 44839-44849
BT-VLoc: Toward Robust and Accurate Visual Localization With Bidirectional Transformer
Visual localization, a fundamental task of numerous computer vision applications, is solved by scene coordinates regression (SCoRe) techniques. These methods leverage neural networks to predict scene coordinates, followed by a PnP algorithm for camera pose recovery. However, they face performance degradation due to ambiguous visual patterns. Recent methods attempt to solve this issue by integrating global and local features, yet the full potential of bidirectional interactions between these features remains underexplored. In this article, we introduce a visual localization framework based on the bidirectional transformer (BFormer), which adaptively models and integrates global and local features through bidirectional interaction. By employing self-modulated convolutions for local feature extraction and self-attention mechanisms in a downsampled space for global information modeling, BFormer enables a dynamic bidirectional adaptation process that strengthens the interplay between local and global representations. As a result, BFormer effectively distinguishes similar image patches by leveraging the complementary strengths of low-level features that provide geometric details and high-level features that contain semantic information. Benefited from BFormer, BTVLoc realizes impressive localization performance on all 7-Scenes, LIVL, and Cambridge datasets.
- Open Access
- 11 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 141765-141780
GCML: Geometric Correlation Encoding Network With Multi-Scale Local Feature Extraction for Accurate Point Cloud Registration
Point cloud registration, a critical component of various applications, aims to establish correspondences between two point clouds. While detector-free methods exhibit outstanding accuracy, they only encode simple geometric features of point clouds while failing to comprehensively model rich geometric correlations between points. Moreover, they treat point clouds as sequential data and only employ the transformer to perform global message propagation during feature enhancement. This approach inevitably neglects the extraction of multi-scale local features, thereby compromising the descriptive capability of point cloud features. This study introduces GCML, a novel detector-free approach that tackles these issues. For the first problem, GCML develops a geometric correlation encoding module (GCEM) that draws inspiration from the Denavit-Hartenberg (DH) modeling method in robotics to effectively encode the geometric correlations between each pair of points within point clouds. This method substantially captures rich geometric structure information inherent in point clouds, thereby improving registration accuracy. For the second problem, GCML proposes a multi-scale local feature perception module (MLFP) that integrates multi-scale local features of point clouds during the feature enhancement stage. Subsequently, GCML employs the transformer architecture to facilitate global message propagation. This design effectively captures both local and global features of point clouds, hence enhancing the descriptive capability of point cloud features. Extensive experiments demonstrate that GCML achieves superior performance on all ModelNet40, DModelNet40, and KITTI datasets.
- 11 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 141765-141780
Open AccessGCML: Geometric Correlation Encoding Network With Multi-Scale Local Feature Extraction for Accurate Point Cloud Registration
Point cloud registration, a critical component of various applications, aims to establish correspondences between two point clouds. While detector-free methods exhibit outstanding accuracy, they only encode simple geometric features of point clouds while failing to comprehensively model rich geometric correlations between points. Moreover, they treat point clouds as sequential data and only employ the transformer to perform global message propagation during feature enhancement. This approach inevitably neglects the extraction of multi-scale local features, thereby compromising the descriptive capability of point cloud features. This study introduces GCML, a novel detector-free approach that tackles these issues. For the first problem, GCML develops a geometric correlation encoding module (GCEM) that draws inspiration from the Denavit-Hartenberg (DH) modeling method in robotics to effectively encode the geometric correlations between each pair of points within point clouds. This method substantially captures rich geometric structure information inherent in point clouds, thereby improving registration accuracy. For the second problem, GCML proposes a multi-scale local feature perception module (MLFP) that integrates multi-scale local features of point clouds during the feature enhancement stage. Subsequently, GCML employs the transformer architecture to facilitate global message propagation. This design effectively captures both local and global features of point clouds, hence enhancing the descriptive capability of point cloud features. Extensive experiments demonstrate that GCML achieves superior performance on all ModelNet40, DModelNet40, and KITTI datasets.
- 22 July 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 205-210
An Improved ICP Algorithm for 3D Point Cloud Registration
The rigid registration of two point clouds is a fundamental task in many areas, such as 3D reconstruction and robot navigation. The Iterative Closest Point (ICP) algorithm has been widely for this task. The basic principle of ICP algorithm is to match corresponding points between the two point clouds and compute an optimal transformation matrix that minimizes the Euclidean distance between corresponding points. A major drawback of the ICP algorithm is the sensitivity to partial overlaps often observed in 3D scans. In this regard, we propose a new rigid registration algorithm based on ICP. Firstly, the Super 4PCS algorithm is used to implement the initial alignment of point clouds to reduce the probability of ICP algorithm falling into a local optimal solution. Then, we propose a local refinement registration method by adaptively eliminating the boundary points of the overlap region of two point clouds. Our algorithm effectively solves the problem that ICP algorithm achieves poor registration accuracy or even produces an incorrect result when dealing with partial overlap data. According to the experimental results, our algorithm achieves better registration accuracy than the classical ICP algorithm.
- 22 July 2022
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 205-210
An Improved ICP Algorithm for 3D Point Cloud Registration
The rigid registration of two point clouds is a fundamental task in many areas, such as 3D reconstruction and robot navigation. The Iterative Closest Point (ICP) algorithm has been widely for this task. The basic principle of ICP algorithm is to match corresponding points between the two point clouds and compute an optimal transformation matrix that minimizes the Euclidean distance between corresponding points. A major drawback of the ICP algorithm is the sensitivity to partial overlaps often observed in 3D scans. In this regard, we propose a new rigid registration algorithm based on ICP. Firstly, the Super 4PCS algorithm is used to implement the initial alignment of point clouds to reduce the probability of ICP algorithm falling into a local optimal solution. Then, we propose a local refinement registration method by adaptively eliminating the boundary points of the overlap region of two point clouds. Our algorithm effectively solves the problem that ICP algorithm achieves poor registration accuracy or even produces an incorrect result when dealing with partial overlap data. According to the experimental results, our algorithm achieves better registration accuracy than the classical ICP algorithm.
- 3 December 2019
- Published by SAGE Publications in Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science
- Vol. 235 (7) , 1193-1210
Design and calibration of a three-dimensional localization system for automatic measurement of long and thin tube based on camera and laser displacement sensor
An integrated system consisting of robot and laser scanner is considered as a promising alternative measurement device for three-dimensional tube which is long and thin. A key to achieving automatic measurement is to locate tube accurately and robustly, which is precondition for scanning path planning. Thus, a novel three-dimensional localization system consisting of camera and laser displacement senor is proposed. When fixed on robot, the camera can search tube quickly in large view and the laser displacement sensor obtains depth information of key points. Measurement principles are presented firstly, including four main steps: camera shooting, key points extraction, laser displacement sensor shooting, and coordinate calculation. Then, the location error influenced by parameters of camera model and laser displacement sensor is analyzed and illustrated by a specific camera and laser displacement sensor. Furthermore, a scan strategy is proposed for quite thin tube localization. Location error caused by perspective is also analyzed and a compensation method is proposed for decreasing the error. Additionally, a sensor transformation calibration method is presented for identifying the relationship between camera and laser displacement sensor, which is verified with high accuracy by calibration experiments. A contrast experiment of a cylinder bar localization shows that the maximum location error of the camera and laser displacement sensor is no more than 3.5 mm, which is only one-eighth of that of Kinect sensor. The mean and maximum location errors are within 2 mm and 3 mm when locating a car brake and fuel tube with about 1300 mm long, indicating high accuracy and good robustness of designed localization system.
- 3 December 2019
- Published by SAGE Publications in Proceedings of the Institution of Mechanical Engineers, Part C: Journal of Mechanical Engineering Science
- Vol. 235 (7) , 1193-1210
Design and calibration of a three-dimensional localization system for automatic measurement of long and thin tube based on camera and laser displacement sensor
An integrated system consisting of robot and laser scanner is considered as a promising alternative measurement device for three-dimensional tube which is long and thin. A key to achieving automatic measurement is to locate tube accurately and robustly, which is precondition for scanning path planning. Thus, a novel three-dimensional localization system consisting of camera and laser displacement senor is proposed. When fixed on robot, the camera can search tube quickly in large view and the laser displacement sensor obtains depth information of key points. Measurement principles are presented firstly, including four main steps: camera shooting, key points extraction, laser displacement sensor shooting, and coordinate calculation. Then, the location error influenced by parameters of camera model and laser displacement sensor is analyzed and illustrated by a specific camera and laser displacement sensor. Furthermore, a scan strategy is proposed for quite thin tube localization. Location error caused by perspective is also analyzed and a compensation method is proposed for decreasing the error. Additionally, a sensor transformation calibration method is presented for identifying the relationship between camera and laser displacement sensor, which is verified with high accuracy by calibration experiments. A contrast experiment of a cylinder bar localization shows that the maximum location error of the camera and laser displacement sensor is no more than 3.5 mm, which is only one-eighth of that of Kinect sensor. The mean and maximum location errors are within 2 mm and 3 mm when locating a car brake and fuel tube with about 1300 mm long, indicating high accuracy and good robustness of designed localization system.