The Activity Fast Recognition of Human in Occlusion and Clusters

This paper proposes an algorithm for human pose discrimination when people are in occlusion and clustering. This algorithm avoids the huge amount of computation loads and speeds up the decision progress. This method is distinct especially when people are in clusters and occluded by each other. For activity discrimination of people, this paper designs a human location feature descriptor-feature of head (FoH), to identify human feature. In this algorithm, human clusters are found, typical single human is then extracted from each cluster by the method proposed by this paper. The center retrieving seed point is proposed which can avoid local extrema, and automatic segmentation method is proposed in this paper. According to the typical single human representation activity, the activity of each cluster can be analyzed with less computation load and the whole activities of the people can be retrieved rapidly. The algorithm is verified by the simulation with the dataset gathered by photos taken from the real environment.