Video Processing And Motion Recognition Model Of Simulated Visual Cortex Neural Network Structure

In recent years,more and more attention has been paid to the research of computer vision,and the action recognition based on video sequence is an important direction in the research of computer vision.Its main purpose is to apply the computer system to automatically analyze and detect the action behavior of human body in the video sequence.Although the research of human action recognition has made great progress,how to extract the action characteristics from video sequences quickly and effectively,and accurately classify the feature information is still an important research topic of computer vision.As a perfect recognition system,the human brain can effectively perceive visual information and accurately recognize human actions.Simulating the mechanism of information processing in the visual cortex of the brain and establishing the brain heuristic human body recognition system is one of the effective ways to solve the problem.Therefore,this paper simulates the properties of the visual cortex neurons,as well as the networks composed of neurons to carries out the research of artificial neural network model.Firstly,the Spatiotemporal information processing model is established by using the properties of the primary visual cortex region(V1)and the middle temporal region(MT)neurons in the dorsal side pathway of the visual cortex.In this model,the temporal and spatial inseparability,velocity selectivity and direction selectivity of neurons in the visual cortex are used to process video information.At the same time,the surrounding inhibition properties of neurons in the visual cortex are used to further process the information,so that the video space-time information can be extracted effectively.Secondly,the pulsed neural network is established by using the biological connection of neurons between V1 and MT.Considering the feedforward interaction between V1 and MT neurons,as well as the directional selectivity of neurons,the mapping relationship between V1 and MT area is established by weighted sampling,and the Pulse neural network is constructed,and the spatiotemporal characteristics of video are effectively extracted.Finally,a human motion recognition method based on LSTM model is presented.By making full use of the associative memory attributes of the LSTM model and the advantages of processing sequence information,the extracted video spatial and temporal characteristics are used as the input of the LSTM model for motion recognition,thus effectively utilizing the time-continuous spatial and temporal characteristics,improving the accuracy of human motion recognition and ensuring the stability of recognition.To demonstrate the effectiveness of the proposed model,this paper chose the KTH human action video database on experimental verification and performance test,the experimental results show that the proposed information processing model,network model and identification methods for human action recognition in video sequences is effective,and the recognition performance than the biomimetic method,or are higher than the biomimetic method.