Fig. 1. Target contour extraction process. (a) Original video; (b) motion detection; (c) target contour extraction

Fig. 2. (a) Template image and (b) test image of POSER 3D simulation samples

Fig. 3. Sampling results

Fig. 4. Distribution of contour points in polar coordinates

Fig. 5. Shape histogram of different contour sampling points. (a) Sampling point 1; (b) sampling point 2; (c) sampling point 3

Fig. 6. Matching results

Fig. 7. Matching results of different sampling points

Fig. 8. Video sequences

Fig. 9. Matching process of feature sequence in spatial domain

Fig. 10. Match process of feature sequence in time domain

Fig. 11. Schematic diagram of DTW algorithm

Fig. 12. Schematic of typical local path constraint

Fig. 13. Diagram of "Morbid" twisting path

Fig. 14. Schematic of parameter of elliptic band

Fig. 15. Recognition rate of different weight distributions

Fig. 16. Influence of decision threshold values on recognition rate

Fig. 17. Schematic of three global constraint boundaries with the same warping window size

Fig. 18. Comparison of searching efficiency of different bands

Fig. 19. Searching efficiency of different global boundaries on large time series

Fig. 20. Classification accuracies of different boundaries sharps and sizes

Fig. 21. Classification accuracies of all warping window sizes with different frames

Fig. 22. Confusion matrix of classification results on KTH dataset. (a) Shape feature; (b) motion feature; (c) fusion feature

Fig. 23. Confusion matrix of classification results on Weizmann dataset. (a) Shape feature; (b) motion feature; (c) fusion feature