Research On Dynamic Gesture Recognition Based On Computer Vision

With the rapid development of computer technology,we have ushered in the age of artificial intelligence,the Internet of Things,and virtual reality.And people are no longer satisfied with the traditional device-centric human-computer interaction methods such as keyboards and mice.Instead,they expect more natural,convenient,user-centric human-computer interaction methods.Gesture recognition has become a hot spot in the field of human-computer interaction because of its advantages such as convenience,naturalness,intuitiveness,and low cost.It can be widely used in VR,smart home,and intelligent transportation.Traditional gesture recognition is based on data gloves.This method uses various sensors to collect gesture data,so it can achieve high recognition rate,but the device is complicated to wear and high in cost,which is not conducive to popularization and use.The gesture recognition based on computer vision adopts the camera to collect gestures,and transmits the collected gestures to the computer for processing.The user also does not need to wear any equipment,so that it has the advantages of convenience,flexibility,and low cost.However,vision-based gesture recognition still faces many difficulties.For example,complex environments,lighting and other factors all increase the difficulty of gesture recognition.Therefore,researchers need to conduct more in-depth research.Dynamic gesture recognition based on computer vision mainly includes four parts:gesture detection and segmentation,gesture tracking,feature extraction and gesture recognition.The research work in this paper is as follows:(1)In the gesture segmentation stage,a method for segmenting a gesture using a combination of skin color detection and background differential is used for the problem that a single method cannot completely and accurately segment a gesture.Firstly,using the clustering of skin color in YCrCb space to detect the skin color in gesture area,face and background.And then using the hand movement characteristics,after analyzing and comparing several segmentation effects based on motion information,the background difference method with simplify and fast is selected to detect motion gestures,it can eliminate the interference of human faces and skin-like objects.The combination of two methods achieved the gesture segmentation.(2)In the gesture tracking phase,taking into account the color characteristics of the gesture,using Camshift algorithm based on the color histogram to track the segmented gestures.For the problem that Camshift algorithm may lead to tracking failure when encounters a lot of skin-color objects,the Kalman filter algorithm is used to predict the position of the gesture movement,so that the combination of the two achieves a good tracking of gestures.(3)In the feature extraction stage,the three most commonly used features of motion trajectories—position,velocity,and direction angle,are chosen to describe the trajectory of the centroid of the gesture,and then use the 8-directional chain code to encode the orientation angle,thus forming a sample of the gesture centroid motion trajectory.Aiming at the problem that the movement trajectory needs to determine the starting point and the ending point,a method is proposed to detect the two states of the open hand and the fist based on the number of convex defect points,finally determines the start and the end of the trajectory.(4)In the recognition phase of hand gesture trajectory,according to the characteristics of space-time difference of gestures,hidden Markov algorithm is used to identify the movement trajectory of the gesture center of mass.Hidden Markov algorithms include Forward algorithm,Baum-Welch algorithm and Viterbi algorithm.The Baum-Welch algorithm was used to train the samples of the centroid motion trajectory of the hand,and then the Viterbi algorithm was used to identify the defined 10 kinds of hand gesture trajectories,and a good recognition rate was obtained.