Multi-Object Tracking (MOT) is one of the core tasks in perception for autonomous driving. It helps vehicles understand the dynamic environment and is useful for subsequent trajectory prediction, motion planning, and other tasks. Most current methods based on LiDAR or camera are not robust enough. LiDAR provides accurate spatial information but suffers when objects are too far away to be detected reliably. Camera-based methods have rich semantic information, but struggle with positioning accuracy. Taking advantage of the rapid development of deep learning in object detection, we propose a novel camera-LiDAR fusion MOT method based on the spatial localization and confidence scores of candidates obtained from both 2D and 3D detectors. Each modality can operate independently in case of sensor failure, and improve its accuracy through the fusion module. Our method also makes the first attempt to associate the cross-modality candidates in only a one-stage association process for better real-time performance. Extensive experiments are conducted on the KITTI tracking benchmark, and our method achieves a competitive performance of 74.04 HOTA on the KITTI test dataset.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    Multi-Object Tracking with Object Candidate Fusion for Camera and LiDAR Data


    Contributors:
    Yin, Huilin (author) / Lu, Yu (author) / Lin, Jia (author) / Schratter, Markus (author) / Watzenig, Daniel (author)


    Publication date :

    2023-09-24


    Size :

    2144018 byte





    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English



    Multiple camera fusion for multi-object tracking

    Dockstader, S.L. / Tekalp, A.M. | IEEE | 2001



    Multiple Camera Fusion for Multi-Object Tracking

    Dockstader, S. L. / Tekalp, A. M. / IEEE Computer Society | British Library Conference Proceedings | 2001


    3D Multi-Object Tracking Based on Radar-Camera Fusion

    Lin, Zihao / Hu, Jianming | IEEE | 2022