Infrared and Laser Engineering, Volume. 51, Issue 10, 20220013(2022)
Target tracking algorithm based on adaptive feature fusion in complex scenes
Figures & Tables(10)
Fig. 4. Comparison of tracking performance of different algorithms
Fig. 5. Success plots of partial interference scenes on OTB100 datasets
Fig. 6. Qualitative evaluation results of 6 tracking algorithms on partial video sequences
Table 1. Improved network structure
View tableView in ArticleTable 1. Improved network structure
Layer Output size Stride Num Input size Conv2 d 3×3 112×112×32 2 1 224×224×3 Block1 112×112×96 1 1 112×112×32 Block2 56×56×144 2 3 112×112×96 Block3 28×28×192 2 3 56×56×144 Block4 14×14×288 2 4 28×28×192 Block5 14×14×384 2 4 14×14×288 Block6 7×7×576 2 3 14×14×384 Block7 7×7×1280 1 2 7×7×576 Avgpooling 1×1×1280 - 1 7×7×1 280 Conv2 d 1×1 k - 1 1×1×1 280
Table 2. Comparison of tracking performance under each improvement strategy
View tableView in ArticleTable 2. Comparison of tracking performance under each improvement strategy
Methods Precision Success rate Network params Proposed 92.53 67.67 $ {\text{3}}{\text{.7}} \times {\text{1}}{{\text{0}}^{\text{6}}} $ ECO-feature+learning_rate 92.30 67.15 $ {\text{138}}{\text{.4}} \times {\text{1}}{{\text{0}}^{\text{6}}} $ ECO-feature 92.14 67.05 $ {\text{138}}{\text{.4}} \times {\text{1}}{{\text{0}}^{\text{6}}} $ ECO-learning_rate 90.91 66.95 $ {\text{138}}{\text{.4}} \times {\text{1}}{{\text{0}}^{\text{6}}} $ ECO 89.90 66.02 $ {\text{138}}{\text{.4}} \times {\text{1}}{{\text{0}}^{\text{6}}} $
Table 3. Comparison of comprehensive performance of different trackers
View tableView in ArticleTable 3. Comparison of comprehensive performance of different trackers
Results Proposed ECO ECO-HC DeepSRDCF SRDCFdecon SRDCF DSST CSK Precision 92.53 89.90 84.76 84.25 81.71 78.11 68.20 51.97 Overlap 67.67 66.02 63.80 63.03 62.20 59.32 51.47 38.49 Mean FPS 15.1 7.1 61.3 1.1 3.7 4.4 50.4 346.9
Table 4. Comparison of precision of each algorithm under different interference
View tableView in ArticleTable 4. Comparison of precision of each algorithm under different interference
Algorithms BC DEF FM IPR IV LR MB OCC OPR OV SV Proposed 95.35 89.02 88.35 90.17 91.35 93.03 90.26 89.84 92.05 87.37 91.57 ECO 86.46 84.93 87.20 87.86 87.46 79.22 86.60 89.61 90.06 83.20 89.53 ECO-HC 84.16 78.20 80.41 78.65 77.38 80.47 78.79 83.05 83.02 81.83 82.45 DeepSRDCF 83.24 75.78 78.89 79.24 74.05 70.22 80.61 80.72 82.11 78.06 82.20 SRDCFdecon 84.12 72.84 75.39 74.66 79.14 67.20 79.92 75.14 78.10 64.07 80.81 SRDCF 76.15 70.84 74.82 71.28 74.29 63.09 75.47 71.82 72.56 59.71 74.93 DSST 69.09 52.88 57.82 68.35 67.51 58.08 58.04 59.48 65.22 47.78 65.42 CSK 57.42 43.62 42.01 53.14 45.07 36.72 38.85 43.08 49.94 31.52 46.33
Tools
Get Citation
Copy Citation Text
Bo Li, Xinyu Zhang. Target tracking algorithm based on adaptive feature fusion in complex scenes[J]. Infrared and Laser Engineering, 2022, 51(10): 20220013
Paper Information
Category: Image processing
Received: Feb. 20, 2022
Accepted: --
Published Online: Jan. 6, 2023
The Author Email:
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20