Research On Human Action Recognition Algorithm Based On Graph Convolutional Networ

Human motion recognition is one of the important application tasks in deep learning.Accurate recognition of the semantic information expressed by human motion can provide convenience for people’s production and life,and is of great significance in all aspects of human life.Most of the existing human motion recognition methods are based on video data and bone point data and adopt deep learning motion recognition methods.These motion recognition methods have achieved very good results at that time,but there are still some deficiencies:On the one hand,there are abundant spatial and temporal features in human bone points,and the model pays insufficient attention to these features;Moreover,the existing methods only extract human motion features on a single scale,and lack the ability to understand actions with different durations.In order to solve the above problems,this paper proposes a human motion recognition model based on multi-scale features.Through the depth feature extraction module and time multi-scale division module,the extraction of hidden features in human bone data is strengthened,and the accuracy of human motion recognition is effectively improved.On the other hand,people need to maintain coordination and balance through the limbs and trunk in the process of movement.Therefore,in some movements,there is a coordination law between the trunk and limbs of the human body;When extracting human motion semantics,due to the natural connection of human body,the model cannot well learn the characteristics of non-adjacent but dependent nodes.In order to effectively solve these problems,this paper proposes a two-stream adaptive multiple attention mechanism graph convolution action recognition model based on action coordination theory.The model can extract the coordination characteristics of human movement,and focus on the information of non-adjacent but dependent nodes.Finally,a large number of comparative experiments are carried out on the public data set in the field of motion recognition to verify the effectiveness of the proposed algorithm.