• Laser & Optoelectronics Progress
  • Vol. 59, Issue 8, 0815002 (2022)
Tong Liu, Sijie Gao*, and Weizhi Nie
Author Affiliations
  • School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China
  • show less
    DOI: 10.3788/LOP202259.0815002 Cite this Article Set citation alerts
    Tong Liu, Sijie Gao, Weizhi Nie. Multitarget Detection Algorithm Based on Multimodal Information Fusion[J]. Laser & Optoelectronics Progress, 2022, 59(8): 0815002 Copy Citation Text show less
    Network structure diagram of the proposed algorithm
    Fig. 1. Network structure diagram of the proposed algorithm
    Pointcloud sampling thresholdConsumed time /msAP /%
    Car EasyCar ModerateCar Hard
    1012460.6455.1349.70
    2017173.2560.4856.28
    3022582.9567.4864.22
    4027781.3468.6165.55
    5032983.2069.1266.37
    Table 1. Performance comparison of proposed algorithm with different pointcloud sampling thresholds on the KITTI validation set
    Grouping methodCarPedestrianCyclist
    W'H'EasyModerateHardEasyModerateHardEasyModerateHard
    200150

    78.42

    82.95

    63.14

    67.48

    61.05

    64.22

    55.43

    58.90

    51.56

    55.33

    47.11

    50.16

    63.79

    68.42

    45.23

    48.52

    44.36

    46.08

    400300
    80060078.8564.2261.5055.8951.9047.6864.2546.0344.88
    Table 2. Comparison of average target detection accuracy of the proposed algorithm on KITTI validation set under different grouping methods
    Fusion strategyCarPedestrianCyclist
    EasyModerateHardEasyModerateHardEasyModerateHard
    Deep fusion78.8563.6261.7954.3151.0547.2164.3846.1443.86
    Late fusion77.4861.7559.6053.0849.8245.1361.7944.6140.22
    Deep & late fusion82.9567.4864.2258.9055.3350.1668.4248.5246.08
    Table 3. Performance comparison of different fusion methods on the KITTI validation set
    MethodmAP /%
    Faster R-CNN69.9
    YOLOv374.4
    Proposed method76.2
    Table 4. mAP comparison of different two-dimensional target detection algorithms on VOC2007 dataset
    MethodModalityCar
    EasyModerateHard
    Mono3DMono2.532.312.31
    3DOPStereo6.555.074.10
    VeloFCNLiDAR15.2013.6615.98
    MV3D(BV+FV)LiDAR71.1956.6055.30
    MV3D(BV+FV+RGB)LiDAR+Mono71.2962.6856.56
    Proposed algorithmLiDAR+RGB82.9567.4864.22
    Table 5. Performance comparison of different methods on the KITTI validation set
    AlgorithmCarPedestrianCyclist
    EasyModerateHardEasyModerateHardEasyModerateHard
    Mono72.2660.4155.8244.7541.3737.9456.4136.9334.65
    Mono+grouping78.7964.0258.6551.8848.1543.2962.1141.2938.90
    Proposed algorithm82.9567.4864.2258.9055.3350.1668.4248.5346.08
    Table 6. Performance comparison of proposed algorithm and the monomodal comparison algorithms on the KITTI validation set
    Tong Liu, Sijie Gao, Weizhi Nie. Multitarget Detection Algorithm Based on Multimodal Information Fusion[J]. Laser & Optoelectronics Progress, 2022, 59(8): 0815002
    Download Citation