Acta Optica Sinica, Volume. 42, Issue 16, 1615001(2022)

Three-Dimensional Multi-Object Tracking Based on Feature Fusion and Similarity Estimation Network

Wenming Chen1,2, Ru Hong1,2, Shaoyan Gai1,2、*, and Feipeng Da1,2,3、**
Author Affiliations
  • 1School of Automation, Southeast University, Nanjing 210096, Jiangsu , China
  • 2Key Laboratory of Measurement and Control of Complex Systems of Engineering, Ministry of Education, Southeast University, Nanjing 210096, Jiangsu , China
  • 3Shenzhen Research Institute, Southeast University, Shenzhen 518063, Guangdong , China
  • show less
    Figures & Tables(15)
    Flow chart of proposed algorithm. (a) Object detection; (b) feature extraction; (c) feature fusion; (d) similarity estimation; (e) data association
    2D feature extraction network model
    3D feature extraction network model
    SE module
    Feature fusion module
    Structure of similarity estimation network
    Flow chart of greedy matching
    Method B in single modal network
    Different methods in multimodal networks. (a) Method C; (b) method D; (c) method E
    HOTA of different modalities and different fusion methods on KITTI verification set
    [in Chinese]
    Bird's eye view of tracking result comparison on verification set. (a) Proposed algorithm; (b) comparison algorithm 1; (c) comparison algorithm 2; (d) comparison algorithm 3; (e) comparison algorithm 4; (f) comparison algorithm 5
    • Table 1. Accuracy of different modalities and different fusion methods on KITTI verification set

      View table

      Table 1. Accuracy of different modalities and different fusion methods on KITTI verification set

      ModalityMethodHOTA /%MAssA /%
      ImageA62.0257.00
      B62.2157.26
      PointA70.8474.15
      B70.9774.41
      Fusion-addC71.2074.83
      D71.3975.28
      E71.5475.55
      Fusion-catC71.2774.98
      D71.4375.36
      E71.6675.79
    • Table 2. Accuracy of different similarity estimation methods on KITTI verification set

      View table

      Table 2. Accuracy of different similarity estimation methods on KITTI verification set

      MethodThresholdHOTA /%MAssA /%
      Euclidean distance0.554.8444.91
      5.055.4146.11
      50.054.8544.63
      Cosine similarity0.552.7341.70
      0.054.2644.15
      -0.552.4241.23
      Similarity estimation network0.571.6675.79
    • Table 3. Accuracies of different algorithms on KITTI test set

      View table

      Table 3. Accuracies of different algorithms on KITTI test set

      AlgorithmHOTA /%MAssA /%MDetA /%
      Complexer-YOLO49.1239.3462.44
      CIWT54.9049.9960.57
      Point3DT57.2059.1555.71
      mmMOT62.0554.0272.29
      Quasi-Dense68.4565.4972.44
      SRK_ODESA68.5165.4975.40
      MOTFusion68.7466.1672.19
      Proposed69.2468.4670.71
    Tools

    Get Citation

    Copy Citation Text

    Wenming Chen, Ru Hong, Shaoyan Gai, Feipeng Da. Three-Dimensional Multi-Object Tracking Based on Feature Fusion and Similarity Estimation Network[J]. Acta Optica Sinica, 2022, 42(16): 1615001

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Machine Vision

    Received: Jan. 13, 2022

    Accepted: Mar. 29, 2022

    Published Online: Aug. 4, 2022

    The Author Email: Gai Shaoyan (qxxymm@163.com), Da Feipeng (dafp@seu.edu.cn)

    DOI:10.3788/AOS202242.1615001

    Topics