Human body behavior recognition method based on depth video sequence

Abstract

The invention discloses a human body behavior recognition method based on a depth video sequence. The method comprises the steps that four-dimensional normal vectors of all pixel points in the video sequence are calculated; low-level features of the pixel points in different layers are extracted by establishing behavior sequence space-time pyramid models in different space-time domains; a group sparse dictionary is learned based on the low-level features, so that sparse codes of the low-level features are obtained; the codes are integrated through a space average pool and a time maximum pool, so that high-level features are obtained and serve as descriptors of a final behavior sequence. The descriptors can effectively preserve space-time multi-resolution information of human body behaviors; meanwhile, the sparse dictionary with stronger expressive power is obtained by eliminating similar contents contained in different behavior categories, so that the behavior recognition rate is effectively increased.

Description

Technical field [0001] The present invention belongs to the technical field of computer pattern recognition, and specifically relates to a method of human behavior recognition based on a deep video sequence. Background technique [0002] Vision is an important way for humans to observe and understand the world. With the continuous improvement of computer processing power, we hope that computers can have part of the human visual function, help or even replace the human eye and brain to observe and perceive external things. With the improvement of computer hardware processing capabilities and the emergence of computer vision technology, people’s expectations of computers may become a reality. Human behavior recognition has always been a research hotspot in the fields of pattern recognition, computer vision, and artificial intelligence. The purpose of video-based human behavior recognition is to understand and recognize individual human actions, the interactive movement between pe...

AI Technical Summary

Problems solved by technology

[0006] The existing human behavior recognition technology based on the normal vector is the most popular. The human behavior recognition method based on the normal vector currently has the following two problems: (1) When using the normal vector to construct the descriptor, since the extraction is based on a certain point The information in the spatial-temporal neighborhood of the layer cannot more accurately represent the detailed information of the local descriptor; (2) when classifying the video sequence of the behavior to be recognized, the atoms in the complete dictionary are used to represent the video sequence to be recognized, if different behaviors The video sequence has similar features, then the similar features are also partially represented, and the incorrectly obtained representation coefficient will reduce the recognition rate for classification

Images