Research On Static Sign Language Recognition Technology Based On Complex Visual Background

Hearing and language disabled people cannot communicate as normal people and their low education level,they cannot express their demands in time most of the time,so the application of sign language recognition is particularly important.In sign language recognition application scenario,the influence of the surrounding environment can bring great challenge to identify and lead to poor effect of sign language recognition under complex background.Sign language includes static sign language and dynamic sign language.Dynamic sign language can be composed of a series of interrelated static sign language.Therefore,static sign language recognition is the basis of dynamic sign language recognition.Current sign language recognition includes two parts,gesture segmentation and recognition.In segmentation,the most common and efficient algorithm is skin color clustering method,but complex background such as skin color will lead to long segmentation time and low recognition rate.To solve this problem,a static sign language recognition research based on vision under complex background is proposed.Firstly,the wavelet transform algorithm is combined with Grab Cut algorithm to segment the image by gesture.Grab Cut algorithm avoids the interference of skin-like background and other objects,and the wavelet transform algorithm shortens the segmentation time.Then,the target sign language is classified through Alexnet convolutional neural network,and the convolutional network model is trained under Caffe framework.A total of 11,400 images were collected,including 30 different types of sign language images.During the signlanguage recognition experiment,3000 images were randomly selected as the training set and the rest as the test set.Experimental results show that this method can achieve a high recognition rate under the influence of light,translation,rotation and complex background.The recognition rate of the method proposed in this paper is 3.7% higher than that of the YCb Cr color clustering method combined with Alexnet,and the segmentation time of the algorithm combining wavelet transform and Grab Cut is 2.032 s shorter than that using Grab Cut alone.