The Research On Object Tracking Of Spatial Context And Color Information Fusion

With the rapid development of intelligent machine in the domestic and foreign, visual information as one of the important human access to information from the outside world, has been focused by more and more computer researchers' attention. In the computer science development process during the past decades, the scientific research aimed at machine vision has become one of very popular field of computer research. The optimization algorithm based on video moving target tracking is one of the most important branches of research in machine vision research. Video tracking has wide application in modern life, Such as intelligent surveillance, traffic control, navigation and military medical diagnosis. So, designing a robust object tracking algorithms has great significance in theory and in practice. In the practical application, tracking moving targets has still existed a variety of challenges, such as the target is partially or completely obscured, the ambient light is changed, the targets' motion is unexpected, the appearance of the target texture is changed, background is mixed, low-resolution, etc. If these negative factors are not addressed, it will adversely affect the effect of target tracking, and even result in tracking failure. So far, there have been many researchers who have made tremendous efforts and contributions to optimize the computer vision target algorithms, but in the unlimited long-time dynamic video, due to the unpredictability of external environment and the target appearance characteristics variability, it's difficult to design a robust tracking algorithm successfully.Currently, video target tracking algorithm based on correlation filters is the most popular tracking algorithm. The idea of correlation filter applied in target tracking is: correlation is a measure of the value of similarity between two signals, the more similar the two signals are, the higher the correlation value is. In designing target algorithm, we first design a filter template, so that when effecting on the target it will produce maximum response, which is the object's position in the current frame. The biggest advantage of the correlation filter is that its speed is faster than others, which is un-comparable by other tracking methods. In this paper, based on correlation filters and aim at the occlusion problem in the target tracking process, we introduced the idea of multi-tracking information fusion, and proposed a target tracking algorithm which combines contextual information and color information to describe the characteristics of the target. The main work in this paper is as follows:1). We have proposed an improved contextual information target tracking algorithm. By improving the confidence map's calculation method (that responds every frame of the target position), the target position and target size scale can be more accurate in the STC target tracking algorithm. And the experimental results also show that the method can effectively improve the accuracy of the original algorithm.2). We have proposed a target tracking method which fuses context information and target information appearance (color information). This method first gets context information by using gray images and target spatial position information between the center and local targets around, and calculates and uses the color information to describe the appearance of the target feature information. Then we design a filter based on a target template based above 2 pieces of information. Finally, by the correlation filter response we obtain the target position in the next frame. And in this way a long robust target tracking algorithm has been realized.The ultimate purpose of this paper is through the above study, to devise an accurate and robust tracking algorithm. The algorithm in this paper has used a variety of characteristic information fusion way to improve the accuracy and success rates, and ensure real-time tracking. In order to verify the practicality and effectiveness of the proposed method, we have made experiments on multiple test video sequences which has been compared with the existing classical tracking algorithms. The experimental results show that our algorithm can handle partially or completely occlusion problems, such as ambient light changes, the scale change and objectives appearance deformation and other abnormal situations. The accuracy and success rates are better than others.