Human movement identification method and equipment, and computer readable storage medium

Abstract

The embodiment of the invention discloses a human movement identification method and equipment, and a computer readable storage medium. The method comprises the following steps that: obtaining a depthvideo image, wherein the depth video image comprises multi-frame depth images; respectively projecting the multi-frame depth images into a space coordinates system to obtain multiple categories of depth sub-image groups; dividing the multi-frame depth images in each category of depth sub-image group into a plurality of sub-movements according to a preset division rule; according to a preset inter-frame difference method, calculating a time depth dynamic model corresponding to each sub-movement; through a spatial pyramid histogram of oriented gradients, coding the time depth dynamic model to generate a feature descriptor; and through a classification algorithm, establishing a corresponding relationship between the feature descriptor and the movement category to identify the human movement.By use of the method, the identification rate of the human movement can be improved, and meanwhile, a human movement identification rate can be improved.

Description

technical field [0001] The present invention relates to the technical field of human motion recognition, in particular to a human motion recognition method and device, and a computer-readable storage medium. Background technique [0002] Human action recognition is an interdisciplinary research topic of image processing, computer vision, pattern recognition, machine learning, artificial intelligence, etc. It has strong practical applications in video surveillance, video retrieval, human-computer interaction, virtual reality, and medical care. sex. [0003] At present, human action recognition methods mainly include: (1) methods based on time-space trajectories, time-space interest points, and time-space silhouettes. However, these methods are based on traditional color images, and these methods are costly and lack two-dimensional information. , the action recognition effect is not good; (2) the action recognition method based on depth data, although the calculation amount o...

AI Technical Summary

Problems solved by technology

[0003] At present, human action recognition methods mainly include: (1) methods based on time-space trajectories, time-space interest points, and time-space silhouettes. However, these methods are based on traditional color images, and such methods are costly and lack two-dimensional information. , the action recognition effect is not good; (2) the action recognition method based on depth data, although the calculation amount of this type of method is small, but this type of method needs to calculate the position of the 3D bone node in an improved spherical coordinate system, but the existing technology The estimation of skeletal nodes is not very accurate, which leads to poor action recognition effect of this type of method

Images