Video-based human body interaction action recognition method

Abstract

The invention discloses a video-based human body interaction action recognition method which comprises the following steps of (S1) carrying out moving target detection on an inputted video frame imageby using an inter-frame difference method, (S2) carrying out feature extraction on a moving target obtained after processing, wherein the step (S2) includes the steps of (S21) extracting human body interaction action features on the moving target obtained after processing by using a mode of combining local space-time features and global optical flow features, (S22) describing optical flow and space-time interest points to form feature descriptors HOF and HOG, and (S23) passing the local space-time features and the global optical flow features through a BP neural network to obtain a probability matrix of an action class under a certain feature, (S3) giving different weights to probability matrixes obtained by using different features and carrying out weighted summation to obtain a fusion probability matrix, wherein an action class with the largest probability is an action class of a frame, and (S4) inputting an initial classification sequence into an improved normal distribution BP neural network to obtain final interaction action classification.

Description

technical field [0001] The invention belongs to the field of computer vision and can be used for human body interaction posture or action analysis research. Background technique [0002] Vision-based human interaction analysis has always been one of the research hotspots in the field of computer vision, which not only has important theoretical significance, but also has broad application prospects. In the intelligent video surveillance system, due to the existence of phenomena such as "robbery" and "fighting", the importance of analyzing and understanding the interaction between people in the scene is particularly prominent; in the huge video database, human interaction Action recognition can automatically retrieve according to predefined action patterns, making it very convenient to retrieve specific events in video databases; virtual reality mainly uses computers to perform visual operations and simulations on complex data to create virtual simulation scenes, and based on ...

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a video-based human interaction action for the existing human interaction action recognition method, which is difficult to effectively extract the features of human interaction actions and establish complex interaction models between multiple targets. Identification technology

Images