Moment Shape Descriptors Applied for Action Recognition in Video Sequences

Algorithms for recognition of human activities have found application in many computer vision systems, for example in visual content analysis approaches and in video surveillance systems, where they can be employed for the recognition of single gestures, simple actions, interactions and even behaviour. In this paper an approach for human action recognition based on shape analysis is presented. Set of binary silhouettes extracted from video sequences representing a person performing an action are used as input data. The developed approach is composed of several algorithms including those for shape representation and matching. It can deal with sequences of different number of frames and none of them has to be removed. The paper provides some initial experimental results on classification using proposed approach and moment shape description algorithms, namely the Zernike Moments, Moment Invariants and Contour Sequence Moments.