Parameter search as a simple computer vision methodology

A novel and simple method is presented to find and identify unique objects in a grey level scene. A database is generated, containing the objects which are expected lo occur in the scene. These objects are stored as lists of consecutive points on an ideal object boundary. An algorithm is developed which maps the stored points onto the input scene, and after each mapping the closeness of the match between the object and the scene is noted. A parameter search is then used to change the mapping until the best match is obtained. The results presented show that the algorithm is very accurate for scenes containing single objects, and for scenes containing multiple objects which do not overlap. For multiple overlapping objects a degradation in performance occurs; however this degradation may be reduced by a more robust cost function. Due to its reliability under certain operating conditions, it is considered that a vision system based on this algorithm would have many applications in highly automated industries where image fields are normally highly constrained or defined.