Gait representation and recognition from temporal co-occurrence of flow fields

摘要: This paper proposes a new gait representation that encodes the dynamics of period through 2D array 17-bin histograms. Every histogram models co-occurrence optical flow states at every pixel normalized template bounds silhouette target subject. Five (up, down, left, right, null) are considered. The first bin counts number frames over in which for corresponding is null. In turn, each remaining 16 bins represents pair and vector has changed from one state to other during period. Experimental results show this significantly more discriminant than previous proposals only consider magnitude instantaneous direction flow, especially as walking gets closer viewing direction, where state-of-the-art recognition methods yield lowest performance. dimensionality reduced principal component analysis. Finally, performed supervised classification by means support machines. using public CMU MoBo AVAMVG datasets proposed approach advantageous methods.