Studies On Target Tracking Algorithms Based On Spatiograms

Target tracking is becoming an active research topic in the areas of computer vision. The essence of target tracking is interactively searching in image sequences to validate the location of a certain object with some salient visual features (such as texture, color, shape, motion). With the rapid development of image processing and the improvement of computer, the technology of target tracking, has been widely used in video surveillance, video retrieval, human-computer interaction, medical diagnosis, traffic control, robot navigation, virtual reality and image guidance and so on.In many target tracking algorithm, because of its real-time and conciseness features, tracking algorithm based on meanshift is widely used. In meanshift tracking framework, histogram is commonly used and is an effective method to describe the characteristics. Many researchers are dedicated to the study of all kinds of histograms to achieve the effective tracking. Color histogram only uses the statistical characteristic of pixels, so when occlusion happens, or background is complex and target goal is small, the algorithm will fail. Spatiogram fuses space information and color information, overcoming the drawbacks of color histogram, so the tracking performance is superior to the color histogram in meanshift framework. But spatiogram needs to do statistic the position of pixels, when rotation, deformation happens to the object, the algorithm is time consuming. Contourlet histogram uses the texture information of the target to build the histogram, and it’s invariance in the main direction, which overcomes the shortcomings of the spatiogram. In this paper, contourlet histogram based with meanshift framework is proposed. The experimental results show that this algrithm can effectively track the target when rotation and deformation happens, and the edge of infrared object can achieve the robust tracking.The actual target tracking often contains kinds of complicated situation, such as small target deform rotation, infrared target occlusion, complicated background happened. To overcome the above issues, feature fusion is one effective methods. In this paper, a tracking algorithm under meanshift framework is proposed which fuses contourtlet histogram and spatiogram, contains the color information, space information of pixels and texture information, uses similarity functions of each frame as weights, then tracks respectively used tracking algorithm based on contourlet histogram and on spatiogram, then add the displacement according to the weights, finally get the tracking results. The experiment results show the proposed algorithm can achieve the robust tracking when rotation or occlusion happens,or when the target is small. Tracking results are more accurate than algorithm based on spatiogram and tracking method based on contourlet histogram.