System Of Human Detection And Tracking Using Kinect

In the field of video surveillance, intelligent analysis of human behavior become the development trend of video monitoring. During this process, recognition and tracking of the human body is of great significance. Through recognition and monitoring of each individual in the video, we can analyze the crowd and the crowd flow mode, so as to extract the effective information for the intelligent video surveillance. In complex scenes, however, due to the influence of the light, the dynamic background, the shelter and cross between two persons and many other factors, human detection failed to get a good result.With the development of the depth camera, and its price also gradually populist, thus, it provides a favorable conditions for the popularity of intelligent video surveillance. In this paper, we use the depth camera Kinect produced by Microsoft to detection and tracking human in complex scenes. With the depth information of the video, some effective information from the depth information has been extracted. Combined with the 2D image information, we are able to identify people from each frame of video better and the recognition accuracy is much higher than just extract information from 2D video data. To track each individual from the detection result, we got the further analysis of monitoring conclusions.To complete the process of this system, the main steps are as follows:Firstly, research the present situation of video monitoring system, the problems which has been already solved and the restrictions still existed at present.Secondly, study the working principle, technical features, interface and application of the Kinect device, etc.Thirdly, the preparatory work of completing the system:the configuration of the Kinect device’s parameters, scene selection and the test to determine the device placement angle, the recording of the system experiment data, etc.Fourthly, based on the experimental data obtained from shoot to complete system:1. Denoise and smooth the original depth image;2. Process distance transform on the depth image, and then, use the shape of Ω to match the head of human preliminarily;3. Using the calibration data, calculate the information of each pixel of the point cloud in the depth image;4. Select the 3D hemisphere for further match of the head of the human body, then we can get the head in the video;5. Separate the human and the ground by the use of regional growth method, then we got the body’s entire profile. The detection has been accomplished.Fifthly, analysis the results of identification and tracking, adjust the parameters, to ensure the system got the optimal configuration.Finally, summarize the characteristics of the system, analysis the problem which the system has already solved and the improvement the system can get, complete the optimization of the system.According to the above requirements, we choose one clothing store in downtown patch as the place for shooting the experimental data. By Using C++ programming, and invoking the openCV library, the system can realize real-time monitor, detection and tracking of human in complex scene.