Hand Gesture Recognition Method Based On Recurrent Three Dimensional Convolutional Neural Network And Attention Mechanism

With the development and advancement of technology,the wave of artificial intelligence has swept the world,making life more intelligent.Among them,humancomputer interaction plays an important role in it.Gesture recognition,as the simplest and most natural way of human-computer interaction,has caused widespread concern.The vision-based gesture recognition does not depend on other devices,and the operation is more natural and convenient.The gesture recognition system in the actual scene usually has high requirements for real-time and accuracy.Therefore,how to improve the real-time performance of the algorithm while ensuring the correctness of the algorithm is the key to the research.Aiming at the difficulty of describing gesture features in video stream and the existence of more redundant information in video sequences,a gesture recognition method based on cyclic three-dimensional convolutional neural network and attention mechanism is proposed.The main work includes:(1)Constructing an end-to-end recurrent threedimensional convolutional nerve,extracting local spatiotemporal features by using threedimensional convolution,and modeling the short-temporal features by using long-and short-term memory networks,which can accurately represent the motion movement in the video.The global spatiotemporal characteristics of the information.(2)There is no dictionary tag for the video sequence,and weighting cannot be performed.A time coding model is defined to segment the video sequence and construct a dictionary tag.(3)Aiming at the problem of redundant information in video,a method of integrating attention mechanism and recurrent three-dimensional convolutional neural network is proposed.The attention mechanism is used to weight the extracted video segmentation features to improve the segmentation.The attention weight of the feature enables the model to pay more attention to important video sequences,thereby improving the accuracy of model recognition.In order to verify the effectiveness of the algorithm,the algorithm was tested on a largescale public dataset and compared with other mainstream methods.It was found that the fusion recurrent three-dimensional convolutional neural network and the gesture recognition method of attention mechanism can not only effectively extract The motion information in the video,and focus on the segments that have a greater impact on the results,and the algorithm is tested in the actual scene,and found to have better real-time and higher accuracy.