Nan Du
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Nan Du has a total of 15 co-authors. Here are the top co-authors by number of shared publications:
- Mu Zhou
Chongqing University of Posts and Telecommunications
7 shared publications - Liangbo Xie
Chongqing University of Posts and Telecommunications
6 shared publications - Jiahao Hu
School of Communications and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing, China
3 shared publications - Jiacheng Wang
Nanyang Technological University
2 shared publications - Rui Xiang
School of Communications and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing, China
2 shared publications - Wei Gao
Department of Computer Science and Technology, Tangshan Normal University, Tangshan, China
2 shared publications
Latest Publications
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 1-13
BioGraphLoc: Bio-Inspired Static Structure Graph for Robust LiDAR Global Localization
Robust and accurate 6-DoF pose measurement in previously mapped environments is a cornerstone of autonomous systems, enabling tasks such as navigation, loop closure, and safety-critical decision-making. In large-scale urban or unstructured settings, Light Detection and Ranging (LiDAR)-based pose measurement has become increasingly vital due to its geometric fidelity and resilience to illumination or appearance changes. This paper presents BioGraphLoc, a novel bio-inspired measurement framework that determines global poses by extracting and encoding structural nodes from a Moving Object Segmentation (MOS)-filtered static structure graph, leveraging biologically inspired mechanisms such as place cells and head-direction cells. Specifically, we first construct a Spatial Static Topology Graph (StatiGraph) from mid-scale static clusters, and encode each node with a lightweight 4D FroZone descriptor that captures depth-level grouping, local geometry, and topology. For reliable correspondence detection, we introduce a Neural Activation Consistency (NAC) strategy and incorporate Hebbian-inspired memory updates to track node stability across time. The final 6-DoF pose is then measured via robust Truncated Least Squares (TLS) optimization with angular regularization. Experiments on the KITTI, HeLiPR, and our custom dataset demonstrate that the proposed method achieves state-of-the-art measurement accuracy, particularly excelling in structurally ambiguous and dynamically cluttered environments where traditional approaches often degrade, with the highest localization success rate reaching 91.32%.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 75 (00189456) , 1-13
BioGraphLoc: Bio-Inspired Static Structure Graph for Robust LiDAR Global Localization
Robust and accurate 6-DoF pose measurement in previously mapped environments is a cornerstone of autonomous systems, enabling tasks such as navigation, loop closure, and safety-critical decision-making. In large-scale urban or unstructured settings, Light Detection and Ranging (LiDAR)-based pose measurement has become increasingly vital due to its geometric fidelity and resilience to illumination or appearance changes. This paper presents BioGraphLoc, a novel bio-inspired measurement framework that determines global poses by extracting and encoding structural nodes from a Moving Object Segmentation (MOS)-filtered static structure graph, leveraging biologically inspired mechanisms such as place cells and head-direction cells. Specifically, we first construct a Spatial Static Topology Graph (StatiGraph) from mid-scale static clusters, and encode each node with a lightweight 4D FroZone descriptor that captures depth-level grouping, local geometry, and topology. For reliable correspondence detection, we introduce a Neural Activation Consistency (NAC) strategy and incorporate Hebbian-inspired memory updates to track node stability across time. The final 6-DoF pose is then measured via robust Truncated Least Squares (TLS) optimization with angular regularization. Experiments on the KITTI, HeLiPR, and our custom dataset demonstrate that the proposed method achieves state-of-the-art measurement accuracy, particularly excelling in structurally ambiguous and dynamically cluttered environments where traditional approaches often degrade, with the highest localization success rate reaching 91.32%.
- 1 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 71 (4) , 9364-9375
PmD-ICP: Enhancing LiDAR Localization Based on Pseudo-Marker Detection and Reflective Marker-Assisted ICP
- 1 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 71 (4) , 9364-9375
PmD-ICP: Enhancing LiDAR Localization Based on Pseudo-Marker Detection and Reflective Marker-Assisted ICP
- 25 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-16
MVD-VE: Motion Variation-Based Dynamic Detection and Velocity Estimation With 3-D LiDAR
With the rapid development of environmental perception technologies and intelligent transportation, accurately detecting and estimating the motion states of dynamic objects has become increasingly important. This study proposes the motion variation-based dynamic detection and velocity estimation (MVD-VE) system, based on 3-D light detection and ranging (LiDAR) point cloud data, aiming for high-precision detection and velocity estimation of dynamic objects. The system first utilizes the multiple signal classification (MUSIC)-based ground normal estimation (MUSIG) for dynamic ground segmentation, effectively eliminating interference and enhancing detection efficiency. Subsequently, a hierarchical detection framework is applied, integrating range image variations and bird’s-eye view (BEV) features, such as vertical robust height range (VRHR) and cell occupancy density (COD), to identify potential dynamic points during coarse detection. A localized-to-global optimization strategy is then used to further refine the detection results, improving accuracy and robustness. For velocity estimation, we propose a robust iterative closest point (R-ICP) algorithm enhanced with adaptive point-pair filtering and termination criteria, addressing outliers and convergence instability. Experiments with a VLP-16 LiDAR show 93.3% peak $F1$ -score in dynamic detection and 0.35-m/s mean absolute error (MAE) in velocity estimation, validating the system’s efficacy for autonomous motion planning and safety-critical applications.
- 25 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-16
MVD-VE: Motion Variation-Based Dynamic Detection and Velocity Estimation With 3-D LiDAR
With the rapid development of environmental perception technologies and intelligent transportation, accurately detecting and estimating the motion states of dynamic objects has become increasingly important. This study proposes the motion variation-based dynamic detection and velocity estimation (MVD-VE) system, based on 3-D light detection and ranging (LiDAR) point cloud data, aiming for high-precision detection and velocity estimation of dynamic objects. The system first utilizes the multiple signal classification (MUSIC)-based ground normal estimation (MUSIG) for dynamic ground segmentation, effectively eliminating interference and enhancing detection efficiency. Subsequently, a hierarchical detection framework is applied, integrating range image variations and bird’s-eye view (BEV) features, such as vertical robust height range (VRHR) and cell occupancy density (COD), to identify potential dynamic points during coarse detection. A localized-to-global optimization strategy is then used to further refine the detection results, improving accuracy and robustness. For velocity estimation, we propose a robust iterative closest point (R-ICP) algorithm enhanced with adaptive point-pair filtering and termination criteria, addressing outliers and convergence instability. Experiments with a VLP-16 LiDAR show 93.3% peak $F1$ -score in dynamic detection and 0.35-m/s mean absolute error (MAE) in velocity estimation, validating the system’s efficacy for autonomous motion planning and safety-critical applications.
- 18 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 75 (2) , 1808-1824
M2CNet: LiDAR 3D Semantic Segmentation Based on Multi-Level Multi-View Cross-Attention Fusion for Autonomous Vehicles
Light Detection and Ranging (LiDAR) point cloud semantic segmentation is an essential task for autonomous vehicles, enabling precise environment perception for safe navigation. Existing point-based and voxel-based methods face limitations related to computational costs and quantization errors, especially dealing with large-scale point clouds. In contrast, 2D projection methods, including Range View and Bird's Eye View, have gained popularity for their balance of real-time performance and accuracy. However, most projection approaches overlook critical issues arising from projection, including 1) inter-class occlusion, 2) boundary-blurring, and 3) inadequate multi-scale feature extraction. To address these issues, we propose a Multi-level Multi-view Cross-attention Fusion Network (M2CNet), which fully leverages complementary information across different views. First, we employ parallel encoder-decoder networks to extract semantic information within a multi-task segmentation framework. Given the high correlation between views, we incorporate a Multi-view Feature Cross Fusion Module (MFCFM) to facilitate cross-view feature association, capturing potential complementarities to mitigate information loss from single-view projections. Then, we place a Multi-cascaded Enhanced Semantic Context Atrous Spatial Pyramid Pooling (MESCASPP) module between the encoder and decoder to capture multi-scale contextual information, enhancing pixel relationships effectively. Finally, the Kernel Point Convolution (KPConv) is employed to refine point-wise features at the output level. The proposed method is evaluated on urban datasets SemanticKITTI and nuScenes (achieving mIoU scores of 65.7% and 78.9% respectively), the off-road dataset RELLIS-3D (44.6%), and a real-world test set. Experimental results demonstrate that our method surpasses representative methods in both effectiveness and scene adaptability. The source code is publicly available at https://github.com/Terminal-lidar/M2CNet.
- 18 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Vehicular Technology
- Vol. 75 (2) , 1808-1824
M2CNet: LiDAR 3D Semantic Segmentation Based on Multi-Level Multi-View Cross-Attention Fusion for Autonomous Vehicles
Light Detection and Ranging (LiDAR) point cloud semantic segmentation is an essential task for autonomous vehicles, enabling precise environment perception for safe navigation. Existing point-based and voxel-based methods face limitations related to computational costs and quantization errors, especially dealing with large-scale point clouds. In contrast, 2D projection methods, including Range View and Bird's Eye View, have gained popularity for their balance of real-time performance and accuracy. However, most projection approaches overlook critical issues arising from projection, including 1) inter-class occlusion, 2) boundary-blurring, and 3) inadequate multi-scale feature extraction. To address these issues, we propose a Multi-level Multi-view Cross-attention Fusion Network (M2CNet), which fully leverages complementary information across different views. First, we employ parallel encoder-decoder networks to extract semantic information within a multi-task segmentation framework. Given the high correlation between views, we incorporate a Multi-view Feature Cross Fusion Module (MFCFM) to facilitate cross-view feature association, capturing potential complementarities to mitigate information loss from single-view projections. Then, we place a Multi-cascaded Enhanced Semantic Context Atrous Spatial Pyramid Pooling (MESCASPP) module between the encoder and decoder to capture multi-scale contextual information, enhancing pixel relationships effectively. Finally, the Kernel Point Convolution (KPConv) is employed to refine point-wise features at the output level. The proposed method is evaluated on urban datasets SemanticKITTI and nuScenes (achieving mIoU scores of 65.7% and 78.9% respectively), the off-road dataset RELLIS-3D (44.6%), and a real-world test set. Experimental results demonstrate that our method surpasses representative methods in both effectiveness and scene adaptability. The source code is publicly available at https://github.com/Terminal-lidar/M2CNet.
- 18 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (13) , 23341-23356
FP-MOS: Frame-to-Frame Prediction for Dynamic Object Segmentation With LiDAR Data
As unmanned robotic operations become prevalent in various fields, the presence of moving objects in complex scenes poses challenges to key technologies, such as environment mapping, obstacle avoidance, and trajectory prediction. In this article, we propose FP-MOS, a LiDAR-based dynamic object segmentation framework that integrates prediction information to continuously capture dynamic object information under constrained data conditions. First, a prediction module (PM) is employed to obtain predicted range images, enhancing the continuity of temporal information in the point cloud within the segmentation network. This improved continuity helps mitigate misjudgment issues caused by dynamic object turns and retrieves previously missed moving objects. Additionally, a motion attention weight guidance module is introduced, which accurately captures dynamic point cloud features through the transmission of point cloud weights between adjacent frames. Finally, we apply an improved adaptive local outlier factor (ALOF) method to filter out outliers in the segmentation results. Experimental results on the SemanticKITTI-MOS and Apollo datasets show that our method achieves leading Intersection-over-Union (IoU) scores of 79.3% and 79.0%, demonstrating the effectiveness of the FP-MOS method in network design and data optimization.
- 18 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 12 (13) , 23341-23356
FP-MOS: Frame-to-Frame Prediction for Dynamic Object Segmentation With LiDAR Data
As unmanned robotic operations become prevalent in various fields, the presence of moving objects in complex scenes poses challenges to key technologies, such as environment mapping, obstacle avoidance, and trajectory prediction. In this article, we propose FP-MOS, a LiDAR-based dynamic object segmentation framework that integrates prediction information to continuously capture dynamic object information under constrained data conditions. First, a prediction module (PM) is employed to obtain predicted range images, enhancing the continuity of temporal information in the point cloud within the segmentation network. This improved continuity helps mitigate misjudgment issues caused by dynamic object turns and retrieves previously missed moving objects. Additionally, a motion attention weight guidance module is introduced, which accurately captures dynamic point cloud features through the transmission of point cloud weights between adjacent frames. Finally, we apply an improved adaptive local outlier factor (ALOF) method to filter out outliers in the segmentation results. Experimental results on the SemanticKITTI-MOS and Apollo datasets show that our method achieves leading Intersection-over-Union (IoU) scores of 79.3% and 79.0%, demonstrating the effectiveness of the FP-MOS method in network design and data optimization.
- 1 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-14
Convex Hull Triangle Mesh-Based Static Mapping in Highly Dynamic Environments
In urban environments, creating an accurate and reliable map is crucial for mobile robotic tasks, including localization, path planning, and navigation. However, the presence of dynamic objects, such as vehicles and pedestrians, can potentially affect static mapping performance. To tackle this challenge, we propose an efficient and robust dynamic removal framework, which establishes a high-quality static map by eliminating interference from dynamic objects. In our framework, a packet-based region growth and secondary fine segmentation (RG-SF) algorithm is proposed to perform adaptive segmentation of the point cloud within a range image, thereby ensuring the accurate removal of dynamic objects. Additionally, we leverage a convex hull triangle mesh (CHTM) to represent the geometric relationships among matched LiDAR points. Furthermore, we utilize the correlation between LiDAR points to distinguish points belonging to static or dynamic scenes by detecting changes in the geometric relationship of adjacent points. Subsequently, dynamic objects are labeled and removed using a multilevel voting strategy, resulting in a comprehensive global static map. Experimental results from SemanticKITTI and our custom dataset showcase that our dynamic removal method significantly enhances static mapping performance and effectively mitigates the influence of dynamic objects across various scenes, surpassing other state-of-the-art methods.
- 1 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-14
Convex Hull Triangle Mesh-Based Static Mapping in Highly Dynamic Environments
In urban environments, creating an accurate and reliable map is crucial for mobile robotic tasks, including localization, path planning, and navigation. However, the presence of dynamic objects, such as vehicles and pedestrians, can potentially affect static mapping performance. To tackle this challenge, we propose an efficient and robust dynamic removal framework, which establishes a high-quality static map by eliminating interference from dynamic objects. In our framework, a packet-based region growth and secondary fine segmentation (RG-SF) algorithm is proposed to perform adaptive segmentation of the point cloud within a range image, thereby ensuring the accurate removal of dynamic objects. Additionally, we leverage a convex hull triangle mesh (CHTM) to represent the geometric relationships among matched LiDAR points. Furthermore, we utilize the correlation between LiDAR points to distinguish points belonging to static or dynamic scenes by detecting changes in the geometric relationship of adjacent points. Subsequently, dynamic objects are labeled and removed using a multilevel voting strategy, resulting in a comprehensive global static map. Experimental results from SemanticKITTI and our custom dataset showcase that our dynamic removal method significantly enhances static mapping performance and effectively mitigates the influence of dynamic objects across various scenes, surpassing other state-of-the-art methods.
- 18 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 24 (14) , 21866-21878
Reflective Markers Assisted Indoor LiDAR Self-Positioning Algorithm Based on Joint Optimization With FIM and GDOP
Simultaneous Localization and Mapping (SLAM) is one of the key technologies in autonomous driving and has been wildly used in mobile robot exploration and unmanned aerial vehicle navigation. LiDAR (Light Detection and Ranging) SLAM depends on matching the data perceived by LiDAR, finding the best match for consistency to determine the LiDAR’s pose, and completing the map construction. However, most existing matching methods are inefficient. Moreover, each processing is based on the previous estimation result, estimation errors accumulate over time, resulting in cumulative drift error. To address issues of low efficiency in extracting key feature points, low self-positioning accuracy and reliability, this paper proposes a LiDAR self-positioning method based on reflective markers (RMs). Firstly, the relationship between the positioning accuracy and the RMs’ coordinates are calculated through Fisher information matrix (FIM). Then, Geometric Dilution of Precision (GDOP) is introduced to comprehensively analyze impacts of the geometric positions and quantity of the RMs on the positioning lower error bound of the LiDAR. The layout rules of RMs are designed according to the vertical angle resolution of the LiDAR resolution. We leverage the high reflectivity of RMs to calculate the LiDAR’s global position. By solving the distance, horizontal angle, and vertical angle information obtained from multiple RMs, high-precision self-positioning is achieved during the LiDAR’s motion. Experimental results show that, compared to the LOAM and LeGO-LOAM algorithms, the LiDAR self-positioning accuracy can be improved by approximately 50% in indoor environments, especially in indoor scenarios involving moving up and down stairs.
- 18 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 24 (14) , 21866-21878
Reflective Markers Assisted Indoor LiDAR Self-Positioning Algorithm Based on Joint Optimization With FIM and GDOP
Simultaneous Localization and Mapping (SLAM) is one of the key technologies in autonomous driving and has been wildly used in mobile robot exploration and unmanned aerial vehicle navigation. LiDAR (Light Detection and Ranging) SLAM depends on matching the data perceived by LiDAR, finding the best match for consistency to determine the LiDAR’s pose, and completing the map construction. However, most existing matching methods are inefficient. Moreover, each processing is based on the previous estimation result, estimation errors accumulate over time, resulting in cumulative drift error. To address issues of low efficiency in extracting key feature points, low self-positioning accuracy and reliability, this paper proposes a LiDAR self-positioning method based on reflective markers (RMs). Firstly, the relationship between the positioning accuracy and the RMs’ coordinates are calculated through Fisher information matrix (FIM). Then, Geometric Dilution of Precision (GDOP) is introduced to comprehensively analyze impacts of the geometric positions and quantity of the RMs on the positioning lower error bound of the LiDAR. The layout rules of RMs are designed according to the vertical angle resolution of the LiDAR resolution. We leverage the high reflectivity of RMs to calculate the LiDAR’s global position. By solving the distance, horizontal angle, and vertical angle information obtained from multiple RMs, high-precision self-positioning is achieved during the LiDAR’s motion. Experimental results show that, compared to the LOAM and LeGO-LOAM algorithms, the LiDAR self-positioning accuracy can be improved by approximately 50% in indoor environments, especially in indoor scenarios involving moving up and down stairs.