Research On Hand Gesture Tracking Based On Adaptive Active Contour Model

With the unceasing development of computer technology, the intelligent controland human-computer interaction based on computer vision have received considerableattention. The new Human Computer Interaction can meet the demand of largequantities of Data Exchange for the people，and has already become one of the futuretendency of the development of the computer. As a convenient and naturalHuman-Computer Interaction mean, Gestures tracking and recognition become moreand more significant.This paper is focused on analysis of gestures tracking technology based onadaptive active contour from the angle of view of the intelligent control andhuman-computer interaction. Gestures detection and segmentation, gestures tracking,model updating and result forecast and adaptive active contour tracking are studieddeeply in this dissertation, of which the main contributions are concluded as follows:1. In view of existing problems of gestures detection inaccurate and gesturessegmentation deviation, a method of gestures detection and segmentation is putforward based on elliptic skin color model in this paper. At first, the skin distributionin YCbCrcolor space will be projected to the CbCrcolor space and the skin color areacan be detected using characteristics of skin projected in the CbCrcolor spaceapproximate elliptic area. Then, the gestures can be detected and segmented by way ofthe method of combining frame difference based on movement. Finally, gesturescontour can be extracted by using the Canny operator. The experimental results showthat the method for gestures segmentation can get better effect during complexsituations.2. To the problem of that the same color in the surrounding environment andsmall movement have great influence on the tracking effect in real-time gesturetracking, a accurate tracking method combining Gaussian mixture model and MeanShift algorithm is proposed in this paper. This method, firstly uses Gaussian mixturemodel to model the surrounding background, and then adopts the background finitedifference method to obtain the gestures image and build gestures’ skin color modelso as to automatic acquisition target tracking, and finally utilizes the Mean Shiftalgorithm to track gotten gestures. Experiments show that in the complex environmentgood results have been achieved for tracking gestures and tracking efficiency has been improved obviously. Especially, it can represent strong stability when the shapes ofgestures change.3. For the problem of that the change of target model can affect the trackingresults in the process of tracking, this paper proposes a method based on Mean Shiftfor updating gesture model and forecasting result. Firstly, it detects and obtainsgestures model through using method combining background difference and skindetection. Then it uses the Mean Shift algorithm to track gestures and update gesturesmodel. Finally, the method taking the Kalman algorithm to predict gestures trackingresults. The Tests results show that the method reduces the effect of the surroundingenvironment on tracking process, and the tracking effect is good.4. For that gestures change and deformation will affect track results in theprocess of tracking, a tracking method using adaptive active contour gesture isproposed based on the level set. This method, firstly uses elliptical skin model andframe difference method to detect and segment gesture, and then uses thesegmentation results to initialize the level set and update it by iteration, and finallyuses Kalman algorithm to follow gestures outline. The experiment shows that themethod can track the outline of gestures better.5. On the basis of VC++6.0, a simple real-time gesture tracking system isdesigned and implemented based on OpenCV. The system can segment gesturesautomatically and select gesture as track target. Continues track of gesture isrelatively stable, and makes preparations for following dynamic gesture recognition.