Skeleton-based shift graph convolutional network human behavior identification method

Abstract

The invention discloses a skeleton-based shift graph convolution network human behavior recognition method, which uses shift graph convolution to reduce parameters of a model and perform dense residual connection on modules, and comprises the following specific steps: step 1, extracting a human skeleton; 2, establishing a network model; the method is composed of nine layers of shift graph convolution modules, and the modules are connected through dense residual errors. Step 3, model learning; and sending human body skeleton data into the constructed network model to start training. And the features are output after passing through a shift graph convolution module, and finally the features are classified through a softmax classifier. The method overcomes the defects that a traditional graph convolution method is large in calculation amount and excessive in model parameters, only physical inherent connection is considered between human skeleton points, and potential connection relations are difficult to excavate, the model parameters are reduced through displacement graph convolution, information of all joint points of a human skeleton is utilized, dense residual connection is adopted between modules, and therefore the method has the advantages that the calculation amount is large; and the classification accuracy is improved.

Description

technical field [0001] The invention belongs to the field of behavior recognition, and in particular relates to a method for human body behavior recognition based on a skeleton by using a shift graph convolution network. Background technique [0002] Action recognition is a relatively active research topic in the field of computer vision. Behavior recognition refers to the creation of algorithms, methods, and frameworks to automatically identify actions performed in videos. The purpose of the human behavior recognition task is to capture and extract the effective behavior information displayed by the human body in the human behavior recognition data, and further determine the type of the behavior. Therefore, human behavior recognition tasks have broad application scenarios. For example, in the field of security monitoring, specific places are monitored through cameras to prevent abnormal behaviors, etc.; Auxiliary help, to deal with abnormal behaviors of patients, reduce t...

AI Technical Summary

Problems solved by technology

Nevertheless, there are still some problems in the human behavior recognition method based on the graph convolutional network based on the skeleton, such as the problem of large amount of calculation, the calculation cost of the network for each action is too high; and the problem of insufficient feature relationship extraction, only considering the The human skeleton is physically inherently connected, and the feature relationship data mining of non-adjacent skeleton nodes is not sufficient

How to better improve performance is still a huge challenge

Images