Research And Implementation Of Video Action Similarity Model Based On Deep Learning

In recent years,with the continuous maturity of hardware technology,the field of human motion recognition analysis has made great progress.With the help of deep learning's powerful feature extraction capabilities,a variety of branches have been developed in the field of human motion recognition,continuously improving the effectiveness of models.Although the research in the field of human recognition has achieved initial results,there is no very effective solution to the problem of human motion similarity analysis.At the same time,in recent years,the demand for rehabilitation diagnosis and treatment has been increasing.Rehabilitation training evaluation is actually a behavior analysis performed by rehabilitation teachers on rehabilitation patients.This topic combines the use of deep learning for human motion similarity analysis and rehabilitation diagnosis and treatment needs,borrows large public data sets,introduces attention mechanisms,constructs human motion image weight masks,and trains a 3D spatio-temporal convolution motion feature extraction model.,And use a suitable similarity measure to compare the similarity of the feature vectors.Finally,a rehabilitation diagnosis and treatment system based on the action similarity evaluation model was constructed to assist rehabilitation patients in rehabilitation training and rehabilitation assessment.This system can greatly reduce the cost of rehabilitation training and alleviate the shortage of resources caused by insufficient rehabilitation talents.The main research work of this paper is divided into the following parts:1.Designed a human motion similarity analysis model based on a 3D Convolutional Neural Network(CNN)and a modified Adjusted Cosine Similarity algorithm that introduced an attention mechanism.Introduce the attention mechanism,combine it with the action feature extraction model,construct a single frame of human contour weight mask in the video data,and homogenize the background instead of the RGB image input of the traditional 3D spatio-temporal convolution model,increasing the complexity of the model Degree to make the feature extraction model pay more attention to the foreground information.The 3D spatio-temporal convolutional network has shown excellent performance in the field of motion recognition.The spatial and temporal features in the motion video sequence are extracted and merged by means of convolution operations.Simultaneously,simple data input and simple network expression are provided.Very excellent operating efficiency.The modified cosine similarity algorithm introduces the idea of centralization and integrates feature vectors into the same feature space to obtain a more objective similarity measure.2.Based on the above-mentioned human motion similarity analysis model,using VueJS+Django+MySQL architecture,combined with motion feature extraction,similarity analysis,motion classification and other algorithm modules,established online rehabilitation video recording,rehabilitation analysis,and online rehabilitation Diagnosis and treatment system.At the same time,the system can save and analyze patient historical data for display.Through this system,it can help patients perform rehabilitation training diagnosis and treatment more conveniently and intuitively.