Based On Deep Learning Research On Action Recognition Of Human Bone Joint Points

In recent years,with the rapid development of information technology and artificial intelligence,the recognition of human action behaviors has been increasingly used in daily production and life fields,and has broad application prospects in the field of human-computer interaction.At present,the traditional detection methods include optical flow method,frame difference method,etc.,but these methods cannot complete the detection and recognition of complex actions,and have low efficiency and high misjudgment rate,which cannot meet the needs of the market.In order to develop a system that can efficiently and intelligently recognize human movements,and to check the standardization of corporate employees ' movements,this project combines the human behavior recognition detection system with deep learning algorithms to build a human behavior recognition system,and verify it through experiments Effectiveness.Aiming at the shortcoming that the existing deep learning methods are very inefficient in extracting the main joint point information of the human body from the image,this paper proposes a method of using the kinect deep infrared camera to scan the human body and transmitting the joint point information matrix to the improved long and short-term memory network.Obtain the input information of the model efficiently and quickly.Compared with the traditional method,it has the advantages of high efficiency,convenience,quickness,and high accuracy.This article mainly analyzes in detail the mechanism of deep learning methods to recognize human actions in detail,and compares the current defects of traditional neural networks in the application of action recognition.Aiming at the problem that the estimated sequence is inconsistent in time caused by the independent error of each frame,a long and short-term memory network model is proposed,in order to enable the long-and short-term memory unit to learn the time dependence of human actions between adjacent frames in the video more effectively,To maintain the time consistency of the action recognition results,propose to use the two-way propagation and attention mechanism to construct the TPA-Bi-LSTM network,and verify the effectiveness of the improved network through experiments,effectively solve the joint point estimation between the two frames The issue of relevance.In order to demonstrate the effectiveness of the method in this paper,the experiment on the host computer deploys the trained model on the C++ platform and writes the relevant interface for testing.By testing and verifying the built system,we optimized the constructed algorithm to the best effect.At the same time,we selected a large number of samples to further test the system.Through the test,we found that the system can accurately identify the behaviors and actions of the human body.The unfinished actions of the human body were judged in time,which further verified the feasibility of the detection method.