Human motion sequence-oriented spatio-temporal pattern analysis is a new problem in pattern recognition. This paper proposes an approach to human motion sequence recognition based on 2D spatiotemporal shape analysis, which is used to identify diving actions. The approach consists of the following main steps. For each image sequence involving human in diving, a simple exemplar-based contour tracking approach is first used to obtain a 2D contour sequence, which is further converted to an associated temporal sequence of shape features. The shape features are the eigenspace-transformed shape contexts and the curvature information. Then, the dissimilarity between two contour sequences is evaluated by fusing the dissimilarity between the associated feature sequences, which is calculated by the Dynamic Time Warping (DTW), and the difference between the pairwise global motion characteristics. Finally, sequence recognition is performed according to a minimum-distance criterion. Experimental results show that high correct recognition ratio can be achieved.
© Copyright 2005 Machine Learning and Data Mining in Pattern Recognition. Published by Springer. All rights reserved.