Motion detection using the randomised Hough transform: exploiting gradient information and detecting multiple moving objects

An algorithm to detect one moving object using the randomised Hough transform (RHT) has previously been proposed. This new nonmodel-based method, called motion detection using randomised Hough transform (MDRHT), was shown be to applicable to translational and rotational motion detection of one moving object. The basic, earlier version of the MDRHT utilises edge points only as its features. The MDRHT is extended to use both edge pixels and the intensity-gradient vector at edge pixels. Moreover, the MDRHT method is generalised to detect also multiple moving objects. The translational-motion experiments with the variant of the technique using gradient information and coping with several moving objects give promising results in two-dimensional-motion detection and estimation, compared with the earlier version of the MDRHT.