• Laser & Optoelectronics Progress
  • Vol. 59, Issue 22, 2215010 (2022)
Zhiling Zhu1, Zhifeng Zhou1、*, Yong Zhao2, Yongquan Wang3, and Liduan Wang3
Author Affiliations
  • 1School of Mechanical and Automotive Engineering, Shanghai University of Engineering Science, Shanghai 201620, China
  • 2Shanxi New Energy Technology Co., Ltd., Taiyuan 030024, Shanxi, China
  • 3Shanghai Compass Satellite Navigation Technology Co., Ltd., Shanghai 201801, China
  • show less
    DOI: 10.3788/LOP202259.2215010 Cite this Article Set citation alerts
    Zhiling Zhu, Zhifeng Zhou, Yong Zhao, Yongquan Wang, Liduan Wang. Multiobject Tracking Algorithm Combining YOLO-V4 and Improved SiameseRPN[J]. Laser & Optoelectronics Progress, 2022, 59(22): 2215010 Copy Citation Text show less
    Overall framework of proposed algorithm
    Fig. 1. Overall framework of proposed algorithm
    Multitarget tracking algorithm flow
    Fig. 2. Multitarget tracking algorithm flow
    Loss and AP curves
    Fig. 3. Loss and AP curves
    Passenger detection results
    Fig. 4. Passenger detection results
    Performance indicator ranking on VOT2016. (a) Accuracy-robustness; (b) average overlap expectation
    Fig. 5. Performance indicator ranking on VOT2016. (a) Accuracy-robustness; (b) average overlap expectation
    Performance indicator ranking on VOT2018. (a) Accuracy-robustness; (b) average overlap expectation
    Fig. 6. Performance indicator ranking on VOT2018. (a) Accuracy-robustness; (b) average overlap expectation
    Comparison of tracking results in different scenarios. (a) Out-of-view; (b) background clutters; (c) partial occlusion; (d) scale variation; (e) deformation
    Fig. 7. Comparison of tracking results in different scenarios. (a) Out-of-view; (b) background clutters; (c) partial occlusion; (d) scale variation; (e) deformation
    Convolution layerConvolution kernelOutput and input channelStrideTemplate imageSearch imageChannel
    127×127255×2553
    Conv13×364×31125×125253×25364
    Res11×13×364×641123×123251×251

    32

    64

    Maxpool12×2261×61125×12564
    Res21×13×3128×64159×59123×12364128
    Res31×13×3128×128157×57121×12164128
    Maxpool22×2228×2860×60128
    Res41×13×3256×128126×2658×58128256
    Res51×13×3256×256124×2456×56128256
    Maxpool32×2212×1228×28256
    Res61×13×3256×256110×1026×26128256
    Res71×13×3512×25618×824×24256512
    Conv21×1256×51218×824×24256
    Conv33×3256×25616×622×22256
    Table 1. Siamese network structure incorporating residual connections
    AlgorithmAccRobEAO
    SiameseFC0.53420.46130.2356
    Staple0.54250.37840.2946
    SRDCF0.53560.41930.2459
    DeepSRDCF0.52710.32640.2758
    TADT0.55390.33270.2991
    SiameseRPN0.56170.26210.3442
    Proposed algorithm0.58080.15610.4095
    Table 2. Comparison results of different algorithms on VOT2016 dataset
    AlgorithmAccRobEAO
    SiameseFC0.51080.48360.2343
    Staple0.52460.68870.1694
    DSiamese0.51170.64580.1966
    SiameseDW0.54110.40320.2704
    CFNet0.50280.58530.1882
    SiameseRPN0.49450.46270.2441
    Proposed algorithm0.53470.34170.3091
    Table 3. Comparison results of different algorithms on VOT2018 dataset
    AlgorithmAUCPrec
    SiameseRPN0.5960.785
    SiameseRPN+RC0.6380.822
    SiameseRPN+AB0.6090.804
    SiameseRPN+RC+AB0.6540.846
    Table 4. Experimental results of ablation on OTB100 dataset
    AttributeNumber of videosSiameseFCStapleSRDCFACFNSiameseRPNProposed algorithm
    SV630.7650.7210.7390.7580.8060.841
    OCC480.7380.7290.7300.7520.7810.837
    DEF430.7790.7420.7250.7640.7930.843
    OV140.6950.6700.5930.6870.7360.780
    BC310.7030.7470.7670.7660.8030.824
    Table 5. Comparison of Prec of different video attributes on OTB100 for each algorithm
    AlgorithmMOTPMOTAFPS
    SORT0.8140.815112
    DeepSort0.8170.86948
    MHT0.8220.8711.5
    POI0.8350.88315
    SiameseCNN0.7640.865107
    Proposed algorithm0.8910.93739
    Table 6. Performance comparison of each algorithm
    Zhiling Zhu, Zhifeng Zhou, Yong Zhao, Yongquan Wang, Liduan Wang. Multiobject Tracking Algorithm Combining YOLO-V4 and Improved SiameseRPN[J]. Laser & Optoelectronics Progress, 2022, 59(22): 2215010
    Download Citation