Better depth images create better motion recognition

Abstract

At present, motion recognition plays an important role in the field of computer vision, and the result of action recognition depends on the extracted feature mode and the expression type of action. Therefore, it is particularly important to propose a relatively fast and robust feature extraction method and apply it to action recognition. In this paper, we propose two feature extraction methods for motion recognition using the information provided by the depth camera, and achieve good recognition accuracy. The two feature extraction methods are as follows: The extraction of temporal and spatial information (ETS) and the clean extraction of temporal and spatial information (CETS). We compare the accuracy of the two feature extraction methods for action recognition, and find that the CETS method has a very good improvement compared with the ETS method. This will have a big significance to our exploration in the aspect of motion recognition.