In this work, a spatio-temporal silhouette representation is put forward, called silhouette principal component image (SPCI), in order to descript motion and shape features for automatic human actionrecognition. SPCI is an image of grey scale and it collects the spatio-temporal sources through emphasizing the temporal variation of different body part. Based on SPCI, we also construct the phase and view variation models. The global shape-based motions descript the spatio-temporal features and variability models.
The construction of optimized model for each action and view is based on the support vectors of motion descriptors from combined action models. In an evaluation of the proposed novel pattern of human action, we achieve high recognition rates on well established benchmark dataset. Our experimental results show that the proposed method of human action recognition is robust and efficient.