Researth On Detection And Tracking Algorithm Of Human Object Based On Depth Image Of Kinect

The technology of object detection and tracking is an important research direction in the field of image processing and pattern recognition. It is widely used in safety monitoring, military reconnaissance, product inspection, human-computer interaction, medical diagnosis and so on. It provides a great convenience for the society. Among them, the detection and tracking technology of human object gradually becomes a research focus at present.The main reason why the robustness of traditional object detection and tracking algorithms based on color image is not high is that when analysing and processing images, affected by illumination change, complex background, object occlusion, shadow and so on. To solve the problem, depth image collected from K inect is used to achieve human object detection and tracking. It not only has the independence of spatial color, but also provides depth information of objects in the scene.This paper uses filtering operation to denoise original depth image. Then an improved mixture Gaussian model based on traditional object detection algorithms is proposed. The algorithm is detailed analysed from modeling, model initialization, model matching to model updating. To solve the cavity and undetected problem caused by slow motion or stationary of human object, this paper uses specific depth information of the image to regulate updating mechanism of the model.For the detected human moving object, firstly, object contour is automatically initialized with Canny operator, and CamShift algorithm of fast operation speed is used to track object. But the anti- interference ability of the algorithm is not high. Thus, we use the method of corner detection to initialize object when interferent appears. Then Snake algorithm based on contour is combined to track object which enhanced the robustness of algorithm. Finally, we display 3D coordinate of human object with depth information. Experiment is simulated under Visual Studio 2010. Results show that the method based on depth image is not affected by interior lighting, shadow and interference. It can detect and track human object accurately and real-timely.