Multiple Hypothesis Tracking in Generalized Hough Transformed Images

In this paper, a new image tracking filter incorporated with GHT (generalized Hough transform) object recognition algorithm is proposed. The GHT process to identify a target shape in the input image is modeled as a Gaussian PDF (probability density function) through in-depth study on the process. The PDF model for GHT is used to systematically determine the target detection threshold for the GHT images in consideration of the target detection probability as well as false alarm probability. The proposed probabilistic GHT process model makes it possible to directly apply the MHT (multiple hypothesis tracking) algorithm to the image tracking problem. The main advantage of the proposed MHT scheme lies with the efficient handling of a priori matching likelihood evaluated from the probabilistic GHT process model. Therefore, it can maintain the target tracking even in the presence of occlusion or clutter around the target-moving path. An illustrative image tracking example is provided to demonstrate the performance of the proposed MHT filter proposed based on the GHT probability model. The example shows that our MHT filter can effectively track a target in clutter environment