Tracking for moving object using invariant moment and particle filter

In this study we used invariant moments with particle filter to track the moving object effectively in video image which is inputted continuously. At first acquiring recursively the mean value and the variance of current each image pixel, then find background image according to above values. Second, subtract adaptively the current image from background image and detect the moving object by using vertical and horizontal histogram. Because the invariant moments is not so sensitive to various transforms such as translation, rotation and scale changes, we can robustly track the moving object by observing particle filter values based on Bayesian probability of pre-distribution and post-distribution. From some experiments, we demonstrate that invariant moment method with particle filter which is suggested above shows better tracking performances than the method of using particle filter only.