Visual feature recognition of human motion in sequence images based on 3D skeleton

The recognition accuracy of human motion visual feature recognition method is low, and the recognition information integrity is low. Based on 3D skeleton, a new visual feature recognition method of human motion in sequence images is proposed. The 3D skeleton information is used to represent human actions, and then human actions are treated as manifolds. Based on 3D skeleton information, main pose features and main action trend features are proposed to represent human actions and measure the distance between actions; Based on the local linearity measurement of human motion manifold, the decomposition of human action is realized, which is conducive to the representation and recognition of classification tasks. The experimental results show that the method based on 3D skeleton can effectively improve the recognition accuracy and integrity, which proves the effectiveness of the proposed method.