Recognition of occluded objects

An effective approach in recognizing occluded objects which are partially blocked from sight is to detect a number of essential features on the boundary of the unknown shapes. Major problems fall into the selection of the appropriate feature set for representing the object in the training stage, as well as in the detection and localization of these features in the recognition process. The method has to be capable of identifying an unknown object based on incomplete feature information, while at the same time be invariant to scale, orientation and minor distortions in boundary shape. Such a scheme is developed and reported in this paper. In this approach, the outermost boundary of an object is transformed into a θ-S domain, and filters with high noise rejection capability are employed to extract the discontinuity points which mark the positions of simple feature segments belonging to the family of circular arc and corners. The detected features are organized into a vector form and classified by the Perceptron to conclude on the identity of the object. An experimental platform is constructed to realize the recognition scheme. The results obtained are satisfactory which demonstrate the feasibility of the approach.