Acta Optica Sinica, Volume. 40, Issue 9, 0915003(2020)

Tracking Algorithm for Siamese Network Based on Target-Aware Feature Selection

Zhiwang Chen1,2, Zhongxin Zhang1、*, Juan Song3, Hongfu Luo1, and Yong Peng4
Author Affiliations
  • 1Key Laboratory of Industrial Computer Control Engineering of Hebei Province, Yanshan University, Qinhuangdao, Hebei 0 66004, China
  • 2National Engineering Research Center for Equipment and Technology of Cold Strip Rolling, Yanshan University, Qinhuangdao, Hebei 0 66004, China
  • 3Jiamusi Electric Power Company, State Grid Heilongjiang Electric Power Co., Ltd., Jiamusi, Heilongjiang 154002, China
  • 4School of Electrical Engineering, Yanshan University, Qinhuangdao, Hebei 0 66004, China
  • show less
    Figures & Tables(15)
    Development process diagram of object tracking algorithm based on Siamese network
    Framework of tracking algorithm with Siamese network based on target-aware feature selection
    Comparison experiment results of original features of each layer in the feature extraction module
    Visualized results of the original features and the target-aware features. (a)(b) Template frame and detection frame; (c)(f)(g) feature maps of layer3; (d)(h)(i) feature maps of layer4; (e)(j)(k) feature maps of layer5
    Visualized results of the original features and target-aware features. (a) Original image; (b) layer3; (c) layer4; (d) layer5
    Specific structure diagram of SiamRPN module
    Visualization of anchor bounding boxes. (a) Visualization of 25×25×K anchor bounding boxes; (b) visualization of 5 anchor bounding boxes at center position of target
    Flow chart of proposed algorithm
    Comparison of success rate and precision plots of OPE for realtime trackers on the OTB100 dataset. (a) Success rate plot;(b) precision plot
    Success rate plots with 11 different attributes for 8 trackers. (a) Low resolution; (b) out of plane rotation; (c) fast motion; (d) out of view; (e) scale variation; (f) motion blur; (g) in-plane rotation; (h) deformation; (i) illumination variation; (j) occlusion; (k) background clutters
    Actual tracking results of each algorithm for videos with different attributes
    • Table 1. Comparison of experimental results of two template frame feature maps on the OTB2015 dataset

      View table

      Table 1. Comparison of experimental results of two template frame feature maps on the OTB2015 dataset

      FrameSuccessPrecision
      Template frame 10.6610.878
      Template frame 20.5470.786
    • Table 2. Comparison of experimental results for changing the number of important feature channels n4 in layer 4 on the OTB2015 dataset (n3=256, n5=1500)

      View table

      Table 2. Comparison of experimental results for changing the number of important feature channels n4 in layer 4 on the OTB2015 dataset (n3=256, n5=1500)

      Video sequenceSuccess, precisionSiamRPN++
      n4=512n4=650n4=800n4=1024
      Basketball0.420,0.5320.432,0.5480.436,0.5570.425,0.5840.446,0.562
      Bird20.708,0.8270.701,0.8250.688,0.8130.698,0.8250.627,0.748
      Bird10.176,0.4970.211,0.5600.203,0.4480.236,0.5020.204,0.367
      Bolt0.261,0.3350.259,0.3370.257,0.3400.650,0.8830.644,0.887
      Girl20.614,0.7200.586,0.6850.557,0.6450.579,0.6790.634,0.720
      Car40.838,0.9530.850,0.9540.849,0.9540.847,0.9510.869,0.953
      ClifBar0.316,0.3960.605,0.8360.577,0.8190.567,0.7950.524,0.718
      Dancer0.779,0.8710.763,0.8520.735,0.8290.741,0.8310.768,0.861
      DragonBaby0.626,0.7480.629,0.7500.630,0.7430.630,0.7470.681,0.830
      FaceOcc10.637,0.5340.636,0.5280.608,0.5600.621,0.5590.604,0.486
      Freeman30.811,0.9540.814,0.9550.816,0.9550.815,0.9540.809,0.960
      Human20.743,0.7730.756,0.7820.768,0.7990.782,0.8060.775,0.817
      Jumping0.550,0.8040.600,0.8400.610,0.8450.578,0.8160.670,0.882
      Liquor0.750,0.8140.708,0.7700.701,0.7640.607,0.6530.616,0.661
      Suv0.745,0.9010.736,0.8980.679,0.8820.434,0.5190.649,0.802
      Woman0.668,0.9010.673,0.9030.670,0.8990.650,0.8900.613,0.906
    • Table 3. Comparison of experimental results on the OTB2015 dataset

      View table

      Table 3. Comparison of experimental results on the OTB2015 dataset

      Tracker nameSuccessPrecisionVFPS
      SiamRPN++0.6950.90535
      Ta-SiamRPN++0.6610.87836
      SiamRPN0.6430.86071
      RASNet[11]0.64283
      SA-Siam[25]0.6570.86550
      CFNet[26]0.5680.74875
      SiamFC0.5820.77149
      TADT0.6470.83934
      DaSiamRPN0.6580.88097
      BACF[27]0.6170.81535
      ECO0.6940.9103
      UPDT0.7020.4
      STRCF[28]0.6833
    • Table 4. Comparison of experimental results on the VOT2018 dataset

      View table

      Table 4. Comparison of experimental results on the VOT2018 dataset

      Tracker nameAccuracyRobustnessEAOLostnumberVFPS
      SiamRPN++0.6010.2340.4155035
      DaSiamRPN0.5860.2760.3835959
      UPDT0.5360.1840.378390.4
      RCO[30]0.5070.1550.376330.8
      DeepSTRCF[31]0.5230.2150.345463
      SA_Siam_R0.5660.2580.3375532
      SiamVGG[32]0. 5310.3180.2866829
      ECO0.4840.2760.280594
      DSiam[33]0.5120.6540.19613810
      SiamFC0.5030.5850.18712532
      DCFNet[34]0.4700.5430.18211627
      DensSiam[35]0.4620.6880.17414719
      Ta-SiamRPN++0.5930.2720.3605836
    Tools

    Get Citation

    Copy Citation Text

    Zhiwang Chen, Zhongxin Zhang, Juan Song, Hongfu Luo, Yong Peng. Tracking Algorithm for Siamese Network Based on Target-Aware Feature Selection[J]. Acta Optica Sinica, 2020, 40(9): 0915003

    Download Citation

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

    Category: Machine Vision

    Received: Dec. 19, 2019

    Accepted: Jan. 19, 2020

    Published Online: May. 6, 2020

    The Author Email: Zhongxin Zhang (ZZXin00016@163.com)

    DOI:10.3788/AOS202040.0915003

    Topics