Moving edge matching for moving object tacking

We propose an edge segment based moving object tracking algorithm using a static camera. The recognition of object from a sequence image is difficult due to the change in object's shape, orientation, motion and size between frames. Objects may contain several parts with motion variation. Moving objects show a wide range of color variation due to the angle of view, illumination change, and reflectance from neighbor objects. Thus, to overcome these limitations, we make efficient use of edge-segments utilizing a Canny edge detector. Moving edge-segments are grouped by means of a iterative k-means clustering algorithm and the group is used in the Generalized Hough Transform based shape matching algorithm due to its robustness to utilize partial information. A Kalman filter is then used to predict the location of each group in future frames. Experiments with outdoor and indoor image sequences show encouraging result under varying illumination conditions with partial occlusion.