Fig. 1. Flow chart of proposed tracking algorithm

Fig. 2. Partial tracking results of single-layer adaptive features and model switching. (a) Skiing; (b) Walking2; (c) Freeman1; (d) Tiger1

Fig. 3. Expression of object characteristics in different convolution layer. (a) Target image; output convolution feature of (b) pool4 and (c) conv5-3 with VGG-19

Fig. 4. Comparison of images (a) before and (b) after adaptive convolution features selection

Fig. 5. PSR and maximum response curves of Freeman1 video tracking process. (a) PSR curves; (b) maximum response curves

Fig. 6. (a) Precision and (b) success rate plots of ten algorithms

Fig. 7. Precision plots of ten algorithms on different challenges. (a) IV; (b) OPR; (c) SV; (d) OCC; (e) DEF; (f) MB; (g) FM; (h) IPR; (i) OV; (j) BC; (k) LR

Fig. 8. Success rate plots of ten algorithms on different challenges. (a) IV; (b) OPR; (c) SV; (d) OCC; (e) DEF; (f) MB; (g) FM; (h) IPR; (i) OV; (j) BC; (k) LR

Fig. 9. Comparison of partial tracking results of six tracing algorithms. (a) Singer2; (b) Tiger1; (c) Car1; (d) Human3; (e) Girl2; (f) Human2