Visual tracking algorithms have important applications in machine vision. Visual tracking algorithms with a moving camera must be robust to camera motion. CAMShift algorithm is a fast and simple color based tracking algorithm but it is not robust to camera motion because it faces with a problem when the tracked object moves across regions of background with similar colors with tracked object. Some algorithms have been proposed to reduce this problem, majority of these algorithms try to reduce the problem by adding more limitations to object model such as adding saturation and value information in addition to hue. By imposing these limitations parts of color similarity problems solved except regions which their colors are similar by hue, saturation and value. Another approach is ABCShift algorithm that tries to reduce this problem by adaptive learning of background. All proposed methods assign a probability of belonging a pixel to the object. In this thesis we assign a probability to regions instead of pixels. Assigning probability to regions leads to more accurate algorithm because we can add some regions information such as area of regions and we can add some information about relation between these regions. We first demonstrate the problems of CAMShift and then we show how this approach can improve CAMShift algorithm especially in situations that there are similar colors in background. Keywords: CAMShift, Region Based CAMShift, RCAMShift, tracking with moving camera
