A complex behavior recognition method

Abstract

The invention discloses a complex behavior recognition method, which includes: using a sensor to obtain three-dimensional bone joint point information of target movement; preprocessing the joint point information and normalizing the coordinate system; extracting the motion track of each joint point, and projecting it to three A two-dimensional plane; extract the motion vector and its length and direction angle between every two frames, use the k-means algorithm to cluster the motion primitives, and obtain the histogram by statistics; use the time pyramid to combine the time information, and combine all the histograms The value of the cluster is used to calculate the weight of each joint point to form a descriptor; the SVM is used to classify to realize action recognition. The present invention can extract and effectively represent the features of action skeleton joint point information, improve the accuracy of action recognition; all motion information is completely preserved, and action reconstruction can be performed; all action classes are clustered to capture human beings from a global perspective Action features; using low-level features reduces the difficulty of calculation, improves the efficiency of action recognition, and meets the real-time requirements of the system.

Description

technical field [0001] The invention relates to a complex behavior recognition method, which belongs to the technical field of image recognition. Background technique [0002] Action recognition is a research hotspot in the field of machine vision. Action recognition methods have been widely used in human-computer interaction, virtual reality, video retrieval, and security monitoring. With the development of depth cameras, the information of human bone joint points can be directly obtained, and the action recognition method based on bone features has greatly improved the accuracy of action recognition. Despite many related studies and exciting results, the effective description of human motion is still a challenging task. [0003] Many methods extract a variety of high-level features from bone information, and then combine them in a certain form to form a descriptor, but the descriptor constructed by this combination mode is not complete, and there is always a loss of motio...

AI Technical Summary

Problems solved by technology

[0003] Many methods extract a variety of high-level features from bone information, and then combine them in a certain form to form a descriptor, but the descriptor constructed by this combination mode is not complete, and there is always a loss of motion information.

On the other hand, many methods train different action classes separately, which leads to bias in the global features of human actions in the description of each individual class action

At the same time, the method of using high-level features still has the problem of high computational cost

Images