Research On People Tracking Based On Kinect

In public and commercial construction section, pedestrian statistics has very important practical significance in security work, scheduling, business arrangement and so on. For the moment, in the field of pedestrian statistics, infrared and photoelectricity have given the predominant place to computer vision. Aiming at adverse light condition, this paper, under the framework of X86 system, utilizes the color and depth image captured by Kinect sensor to track people and get the accurate statistical data of people number. This thesis has positive significance in science and engineering.Firstly, the operation principle of Kinect and the reason of distortion are anglyzed. MATLAB Kinect calibration toolbox is used to calibrate the Kinect sensor. This process will make the depth image and the color image match better. Then, pixels in empty holes’ neighborhood region are used to restore the depth information. The method gets good result.After the fixed depth image is acquired, Mean Shift segmentation method is used to find the heads in the images. Considering that the Mean Shift may take too long time to segment the image, we put forward a method, based on watershed algorithm, to strengthen the timeliness. In view of the situation that the Kinect cannot get depth information well under strong sunshine, we use HOG operator and support vector machine(SVM) to detect people in the color image. Experimental results show that the proposed method has good performance.Mean Shift algorithm and Kalman filter are two of the most effective methods in the field of video target tracking. Mean shift tracking method cannot track the object move fast in the image, while Kalman filter will predict the position of the moving object approximately. Therefore, we use Kalman filter to predict people’s trace when they are detected. Then, Mean Shift algorithm is used to acquire more accurate positon of people who are tracked and counted. Experimental result demonstrates that the proposed method can work well under a variety of lighting conditions. People walk in a normal way will be tracked and counted accurately under normal lighting condition. Under strong light condition or people pass the gate in an abnormal way, for example, carrying a case, the accuracy will decline to some extent.Finally, the research work is summarized. Future study is also pointed out.