Research Of Dynamic Gesture Recognition Method Based On Neural Network

With the continuous development of artificial intelligence,gesture recognition algorithms have been widely used in the fields of deaf-mute assistance,virtual augmented reality,and automotive vehicle systems.Gesture recognition technology has become an important topic in the field of human-computer interaction.At present,vision-based dynamic gesture recognition is mainly faced with the problem of interference information under complex dynamic backgrounds,redundant gestures,similar feature information and continuous contact.In response to the problems in gesture recognition,this paper uses a dynamic gesture recognition method based on neural network to recognize dynamic gestures.The main research contents of the paper are as follows:(1)In view of the complex background interference information,this article will grayscale the frame sequence of dynamic samples,use median filtering and threshold noise reduction to perform noise reduction on image samples,and finally use image binarization to make the entire image appear obvious.Black and white effect,conducive to the next detection and identification.(2)For the recognition of dynamic gestures,this paper uses the 3D convolutional neural network model and uses the 3D convolutional layer and 3D pooling layer to effectively extract the short-temporal feature expression of dynamic samples.Subsequently,the output features of the 3D convolutional neural network are used as input to the two-way convolutional long-term and short-term memory network.Under the condition that the spatial features are not affected,the long-temporal characteristics of gestures are further extracted to enhance the continuity and Contact to discover the meaning of continuous action information.Finally,in order to ensure the portability of the model and the efficiency of operation,this paper proposes to assign a corresponding weight factor to each frame in the sequence by adding a spatial attention mechanism,adding a time attention mechanism to indicate the time point of important features,and using attention The mechanism filters feature information,reduces the amount of redundant calculations,improves the utilization rate of computing resources,and improves the extraction efficiency of feature information.(3)The thesis experiment is implemented using TensorFlow,and a recognition method combining a 3D convolutional neural network module,a two-way convolutional long and short-term memory network and an attention mechanism module is proposed,and classification performance is performed on the dynamic gesture data set and the Chinese traffic police gesture data set.experiment.Through comparison and control experiments with other algorithms.Experiments show that the recognition method in this paper has a good effect on simple dynamic gesture recognition,and has a good recognition rate and operating efficiency.