Research On Gesture Recognition Technology Based On Computer Vision

With the rapid development of visual technology and artificial intelligence,people have higher and higher requirements for human-computer interaction.The human-computer interaction method based on human biological characteristics provides a more convenient and flexible method.As the second language of human beings,gestures have rich meanings.Gesture recognition is a research hotspot in human-computer interaction.Among them,because the gesture recognition technology based on computer vision has the characteristics of natural and friendly user-oriented operation,it has become a research focus in the field of gesture recognition.However,hand gestures have certain complexity and are greatly affected by the background environment,which leads to many problems in gesture recognition.Therefore,this paper has conducted in-depth research on key technologies such as gesture segmentation,hand feature extraction and gesture recognition,and solved the above problems to a certain extent.After researching and comparing the effect of skin color segmentation effect in different color space,this paper selects YCbCr color space as the skin color space,and use the skin color segmentation algorithm based on the ellipse skin color model to segment the skin area of the hand;then,the cascade classifier based on Haar features is used to quickly locate the hand area;finally,the skin color segmentation and the cascade classifier based on Haar features are combined to segment the gesture area.This algorithm can effectively avoid the interference of skin-like objects,ensure the integrity of gesture segmentation,and maintain a good segmentation effect in complex environments.In the hand feature extraction,this paper extracts the HOG feature of the gesture image to characterize the global feature of the gestures,and at the same time improves the fingertip detection algorithm based on the center of gravity to ensure the accurate and rapid extraction of the gesture fingertip feature.Then,the HOG features that have undergone PCA dimensionality and the fingertip features are serially weighted and fused,and finally the feature vector of the gesture is obtained.In the process of gesture recognition,this paper builds a DAGSVM multi-classification model based on directed acyclic graph to classify gestures.After extracting the feature vector of the gesture and sending it to the classifier for training,the trained classifier can be used to realize gesture recognition.Through multiple sets of experiments on sample set,it indicated that the algorithm in this paper can recognize different gestures more accurately,and the algorithm has better stability under different lighting conditions and different background environments.