Author Affiliations
1 Fujian Institute of Research on the Structure of Matter, Chinese Academy of Sciences, Fuzhou, Fujian 350002, China2 School of Data Science, North University of China, Taiyuan, Shanxi 0 30051, China3 College of Electrical and Electronic Engineering, Shandong University of Technology, Zibo, Shandong 255049, Chinashow less
Fig. 1. Framework of amStaple algorithm
Fig. 2. Tracking results of all thresholds ε on different benchmarks. (a) Precision curves; (b) success rate curves
Fig. 3. Tracking results of nine algorithms on different benchmarks. (a) Precision for OTB-2013 benchmark; (b) success rates for OTB-2013 benchmark; (c) precision for OTB-100 benchmark; (d) success rates for OTB-100 benchmark
Fig. 4. Seven attributes in OTB-100 benchmark sequence where amStaple has best performance. (a) Illumination variation; (b) out-of-plane rotation; (c) occlusion; (d) deformation; (e) in-plane rotation; (f) out-of-view; (g) background clutter
Fig. 5. Qualitative comparison of top three algorithms on ten videos. (a) Bird1; (b) Bolt; (c) ClifBar; (d) coke; (e) dog; (f) dragon baby; (g) shaking; (h) girl; (i) soccer; (j) trellis
Fig. 6. Video attribute success rate curves where amStaple1 performs better than amStaple on different benchmarks. (a) Fast motion for OTB-2013 benchmark; (b) motion blur for OTB-2013 benchmark; (c) low resolution for OTB-2013 benchmark; (d) fast motion for OTB-100 benchmark; (e) motion blur for OTB-100 benchmark; (f) low resolution for OTB-100 benchmark
Fig. 7. Tracking effects of amStaple1 and amStaple in Deer benchmark sequence. Red is amStaple and green is amStaple1
Parameter | Value |
---|
HOG cell size | 4×4 | HOG orientations | 9 | Learning rate (correlation filering) ηcf | 0.01 | #bins colour histograms nhist | 25×25×25 | Hyperparameter δ | 2.22×10-16 | Hyperparameter ξ | 2 | Threshold θ | 0.5 |
|
Table 1. Basic experimental parameter settings
Benchmark | Criterion | Staple | amStaple | Increase ratio /% |
---|
OTB-2013 | Precision | 0.782 | 0.833 | 6.52 | Success rate | 0.593 | 0.622 | 4.89 | OTB-100 | Precision | 0.784 | 0.810 | 3.32 | Success rate | 0.579 | 0.597 | 3.11 |
|
Table 2. Precision and success rates of amStaple and Staple on different benchmarks and percentage increases of amStaple compared to Staple
Attribute | Successof Staple | Success ofamStaple | Increaseratio /% |
---|
Illumination variation | 0.595 | 0.628 | 5.55 | Out-of-plane rotation | 0.534 | 0.555 | 3.93 | Scale variation | 0.520 | 0.546 | 5.00 | Occlusion | 0.543 | 0.563 | 3.68 | Deformation | 0.550 | 0.559 | 1.64 | Motion blur | 0.540 | 0.558 | 3.33 | Fast motion | 0.541 | 0.557 | 2.96 | In-plane rotation | 0.549 | 0.573 | 4.37 | Out-of-view | 0.476 | 0.507 | 6.51 | Background clutter | 0.561 | 0.601 | 7.13 | Low resolution | 0.399 | 0.405 | 1.50 |
|
Table 3. Comparison of success rates of amStaple and Staple on different attributes of OTB-100 benchmark