Author Affiliations
1College of Computer Science, Sichuan Normal University, Chengdu, Sichuan 610101, China2Faculty of Electrical Engineering and Computer Science, Ningbo University, Ningbo, Zhejiang 315211, Chinashow less
Fig. 1. Framework of proposed algorithm
Fig. 2. Target-feature visualization. (a) Original image; (b) HOG feature; (c) conv3-4 feature; (d) conv4-4 feature; (e) conv5-4 feature
Fig. 3. Schematic of feature fusion
Fig. 4. Update mechanism of temporal model
Fig. 5. Flow chart of algorithm
Fig. 6. Qualitative results of 10 tracking algorithms for some video sequences
Fig. 7. Tracking results of different feature experts
Fig. 8. Tracking accuracy and success rate of algorithm on OTB-2013 and OTB-2015 databases. (a) Tracking accuracy on OTB-2013 database; (b) tracking success rate on OTB-2013 database; (c) tracking accuracy on OTB-2015 database; (d) tracking success rate on OTB-2015 database
Fig. 9. Tracking precision of 11 different attribute video sequences on OTB-2013 database. (a) Background clutter; (b) deformation; (c) fast motion; (d) in-plane rotation; (e) illumination variation; (f) low resolution; (g) motion blur; (h) occlusion; (i) out-of-plane rotation; (j) out of view; (k) scale variation
Fig. 10. Tracking success rates of 11 different attribute video sequences on OTB-2013 database. (a) Background clutter; (b) deformation; (c) fast motion; (d) in-plane rotation; (e) illumination variation; (f) low resolution; (g) motion blur; (h) occlusion; (i) out-of-plane rotation; (j) out of view; (k) scale variation
Fig. 11. Tracking precision of 11 different attribute video sequences on OTB-2015 database. (a) Background clutter; (b) deformation; (c) fast motion; (d) in-plane rotation; (e) illumination variation; (f) low resolution; (g) motion blur; (h) occlusion; (i) out-of-plane rotation; (j) out of view; (k) scale variation
Fig. 12. Tracking success rates of 11 different attribute video sequences on OTB-2015 database. (a) Background clutter; (b) deformation; (c) fast motion; (d) in-plane rotation; (e) illumination variation; (f) low resolution; (g) motion blur; (h) occlusion; (i) out-of-plane rotation; (j) out of view; (k) scale variation
Fig. 13. Tracking precision and success rate of algorithm under long-term tracking. (a) Tracking precision;(b) tracking success rate
Expert | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 |
---|
Frequency | 18 | 54 | 18 | 80 | 77 | 85 | 25 | 34 |
|
Table 1. Frequency statistics of feature experts
Database | Parameter | Ours | STRCF | MCCT | CF2 | ECO | SRDCF | ADNet | Staple | KCF | LCT |
---|
OTB-2013 | Precision | 0.902 | 0.889 | 0.883 | 0.891 | 0.855 | 0.838 | 0.798 | 0.782 | 0.740 | 0.848 | | Success rate | 0.876 | 0.845 | 0.863 | 0.809 | 0.806 | 0.789 | 0.721 | 0.738 | 0.623 | 0.738 | OTB-2015 | Precision | 0.871 | 0.864 | 0.860 | 0.845 | 0.836 | 0.788 | 0.772 | 0.784 | 0.696 | 0.762 | | Success rate | 0.829 | 0.800 | 0.818 | 0.751 | 0.772 | 0.730 | 0.700 | 0.699 | 0.526 | 0.629 | Average FPS | 3.9 | 28.0 | 4.2 | 1.5 | 54.8 | 8.0 | 12.3 | 97.6 | 349.3 | 29.4 |
|
Table 2. Tracking accuracy, success rate, and speed of algorithm on OTB-2013 and OTB2015 databases