Research On Chinese Sign Language Recognition Technology Based On Computer Vision

Sign language is an important medium for deaf people to communicate with the outside world.Research on Chinese standard sign language recognition based on Putonghua can facilitate communication between deaf and normal people,and also promote the development of human-computer interaction.In recent years,with the rapid development of computer technology and deep learning,the use of computer vision sign language or gestures for classification and recognition has gradually become a research hotspot for researchers at home and abroad.The main research content of this paper is the recognition of Chinese finger language based on computer vision,including the establishment of a finger language image database,preliminary preprocessing,feature extraction and classification recognition of finger language images.details as follows:Firstly,the finger language images are collected to establish a finger language image database,and the finger language images in the image database are pre-processed.The preprocessing operations include skin color clustering,image smoothing,morphological transformation,and image segmentation.Finally,the images are uniformly sized as 300×300.The preprocessing operation is the basis for subsequent feature extraction and recognition of finger language images.Secondly,aiming at the problem of selecting the optimal parameters in the process of SVM model construction,the gray wolf optimization algorithm was introduced to optimize the SVM parameters to realize the recognition of finger language.Analyze common feature extraction methods and classifiers,and select HOG features and SVM classifiers.The gray wolf optimization algorithm is used to optimize the SVM parameters.The time to find the optimal parameters is shorter and the recognition speed is faster.In finger language recognition,HOG features are extracted,SVM parameters are optimized using the gray wolf optimization algorithm,and sent to the radial basis function(RBF)SVM classifier to achieve finger language recognition.Experiments show that this algorithm can be used for finger language recognition and has certain effectiveness and feasibility.Finally,for traditional manual manual feature extraction,subjectivity is strong,and other deep learning requires higher data sets.a deep transfer learning algorithm is used to realize finger language recognition.The effects of Inception V3 and Res Net50 models applied to replace only the classifier's parameter migration and freezing different network layer fine-tuning models on finger language recognition are compared.Experiments show that the Res Net50 model has better recognition effect for finger language.The Res Net50 model was used to freeze the 50 layers of the network layer and fine-tune other parameters to predict the sign language image.Experiments show that the deep transfer learning algorithm used in this paper can be used for finger language recognition,and has certain advantages compared with traditional artificial feature extraction features and other deep learning algorithms.