The Research Of Unspecified Dynamic Sign Language Recognition Based On Video

With the rapid development of computer technology,the "human-centered" human-computer interaction has become the focus of current research.In particular,research based on computer vision gesture recognition has attracted more and more attention due to its natural habits and low equipment costs.At the same time,sign language is an important way for deaf and dumb people to use hand-type information instead of sound language to communicate.Since healthy people generally do not understand sign language and sign language translation belongs to emerging professions,they are still unable to meet market demands,so the study of sign language recognition is very meaningful.To meet the needs of the sign language/voice two-way recognition system in the smart community in the future and to achieve visual "talking" between deaf and dumb people,it is necessary not only to establish sign language recognition for non-specific people,but also to meet the real-time requirements of the system..In order to effectively solve these problems,the existing algorithms in the identification of opponent names are studied and improved in depth,and the following two aspects are mainly completed.1)A key frame extraction algorithm based on compressed sensing and SURF feature is proposed to recognize the real-time,large vocabulary sets and continuous sign language videos efficiently and accurately in this paper.Using compressed sensing,the gesture video is reduced to low dimensional and multi-scale frame image features.The segmentation of sub lens is done by adaptive threshold method,and a large number of hand gesture frame data are processed;SURF feature points are usedto complete feature matching,and the SURF features' similarity curve in frames is drawn to extract the key frames.In the pre-processing stage,HSV space adaptive color detection is used to abstract the gesture area.Experimental results show that the algorithm has the ability to deal with a large number of complex data,and the recognition results are more accurate.2)Aiming at the problem that TLD algorithm can not track failure under the conditions of uneven illumination variation,occlusion and tracking target ambiguity,a TLD tracking algorithm based on convolution neural network is proposed.Firstly,the gesture is taken as the positive sample and the background is taken as the negative sample.The HOG features extracted from the gesture are input into the convolution neural network for training,and the gesture detection classifier is obtained,and the target area is determined,the gesture is automatically identified,and then the gesture is performed by using the TLD algorithm Tracking and learning,it is estimated that positive and negative samples are detected and corrected in real time,while tracking data is updated.Experiments show that the tracking accuracy of the algorithm is higher than that of the traditional TLD algorithm,and has higher robustness.