A Video Gesture Processing Method Based on Convolution and Long Short-Term Memory Network
Xiaoxue Ding, Chao Xu, Quya Yan · 2019
As there are many deaf-mute groups in today's society, it is particularly important to solve the problem of communication barriers. In view of the limitations of wearing equipment and the low recognition rate of non-equipment at the present stage, using the network to extract hand information and using different features and classification methods to recognize dynamic hand information has become the mainstream. In this paper, Optical flow method is used to detect and process moving target objects on video. The method of combining Convolution Neural Network with Long Short-Term Memory (CNN + LSTM ) was improved and compared with Convolution Neural Network and Support Vector Machine (CNN + SVM ). Experiments show that the optical flow method is used to process video motion information, which not only contains the description of the motion information of moving objects, but also the information of the three-dimensional structure of the scene. After the processed video stream is trained by CNN + SVM , CNN +LSTM and the improved CNN + LSTM , CNN + SVM has a lower recognition rate than CNN + LSTM, The improved network uses PReLU as activation function to improve the gate structure of LSTM, and The recognition accuracy increased from 61.3% to 92.4%.