View angle-independent action identification method

Abstract

The invention discloses a method for identifying an action irrelevant to visual angles. The method is executed according to the following steps: first, synchronously collecting human body video data in the forward direction, an oblique direction and a side direction, pre-treating the collected video data so as to obtain a binary human body profile information, and carving and reconstructing a three-dimensional figure of the human body according to the binary human body profile information in the three directions; then extracting the dynamic part of the motion process of the human body to form a motion dynamic energy body and a motion weight model of the three-dimensional figure, and performing characteristic description by utilizing a three-dimensional Zernike torque without changing proportion, displacement and rotation; finally using a conditional random field to establish a probability graph model for each action and performing identification. The method for identifying actions irrelevant to visual angles overcomes the restriction that an existing monitoring scene requires a fixed visual angle, and realizes the identification of human body actions in any directions, and the identification rate is high.

Description

technical field [0001] The invention belongs to the technical field of intelligent visual monitoring, and in particular relates to an action recognition method irrelevant to an angle of view. Background technique [0002] With the rapid development of science and technology, the use of cameras to monitor scenes has been widely used in all aspects of society. The main target in the monitoring scene is people, and the recognition of various actions of the human body helps to prevent and stop crimes. So far, most research work on action recognition has been carried out under a fixed perspective. Although a small number of viewpoint-invariant representations are used for research, most of them have some defects, such as lack of sufficient information for recognition, relying on robust semantic feature point detection or point correspondence. Since the angle between the direction of human body movement in the scene and the shooting direction of the camera is arbitrary, it is im...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a view-independent action recognition method, which solves the limitation of fixed view angle in the existing monitoring scene, realizes human action recognition in any direction, and has a high recognition rate

Images