Study On Gesture Tracking And Recognition Technology For Human Computer Interaction

Human-computer interaction mainly refers to the information interaction between human and intelligent machines.The transform from mouse control to multi touch control,and to somatosensory technology,the continue progress subverts the traditional ways of interaction.In recent years,with the development of deep learning,pattern recognition and intelligent information processing technology,Human-computer interaction technology has been a gradual transition to computer vision,which is considered to be in a more natural way to identify human behavior and brings a new experience to the users.Gesture interaction is one of the simplest and directest human-computer interactions.It consists of detecting the gesture target from the input video sequence and tracking the gesture,further recognizing the tracked gesture,and finally translating the recognition result into a language which machine can understand.This paper starts from the estimation of nonlinear system,divide the man-machine interaction system into three modules,and the different modules are cooperated with each other to achieve a high reliability application system.The three core modules are palm gesture detection module,deformation gesture tracking module and static gesture recognition module.The detection module is used to start the entire interactive system by senting the detected palm gesture position to the tracking module.In this module,the skin color model and the gradient histogram feature are combined to achieve a fast and accurate palm gesture detection.Deformation gesture tracking module is the key link of Human-computer interaction technology based on the behavior of the gesture,in which the biggest difficulty will be the irregular shape and the uncertainty trajectory of the target gesture,as well as the high requirement for real-time.In view of the limitations and constraints of the existing gesture tracking algorithms,a nonparametric estimation method based on kernel density function is adopted in the tracking module.The advantage of the tracking module is that it does not require prior knowledge but relies on the existing data to estimate.It can be used to estimate the probability density of arbitrary shape.The probability density map of each frame is obtained by estimating the posterior probability density using the target modeling.Then,the probability density map is decomposed into the gesture motion area and the color interference region.In order to weaken the interference of similar color objects,the color interference region is subjected to motion modeling base d on the Gaussian distribution.The redetection module is designed to solve the problem of target missing.The results show that the tracking module in this paper can capture the non-linear motion of the gesture rapidly when the appearance of the gesture changes drastically.In the static gesture recognition module,the convolutional neural network and the K-Nearest Neighbor classifier are used to recognize the tracked gesture,the recognition of seven kinds of gestures is realized in the end.Experimental results show that the cooperation between recognition module and tracking module completely meets the real-time requirement of the system,and achieves a high stability of the human-computer interaction in mutiple scenarios.