Recognition of sports action pattern based on sparse matrix vector

摘要

The Thesis proposes an algorithm of physical gesture recognition of human body based on composite features, which combines the feature of contour vector distance from the contour mass center to the edge point with the optical-flow features, and then the composite features are formed to make action recognition. According to the recognition results in the third part, we can see that such composite features achieve more than 95% correct recognition rate in the Weizmann database and KTH database, which proves the validity and feasibility of such algorithm sufficiently.