A vision system for automatic object recognition

This paper presents the results of the integration of a proposed vision system based on the decomposition of objects into parts. The objects are defined by their contours which are obtained using a hybrid first and second-order differential edge detector. The parts are isolated, and their parameters modelled by 2-D superquadrics, are obtained by the minimization of a best-fit cost function. The object is then represented by its structural description which are data structure whose predicates represent the parts and whose arguments represent the spatial relationship between the parts. This representation allows the recognition of objects independently of their position, orientation or size and is insensitive to partially missing parts. In this paper, examples illustrating acquired images of objects, the extraction of their contours, the isolation of the parts, and their fitting with 2-D superquadrics are reported. The recognition rate of the suggested vision system for simultaneously and randomly rotated, displaced and resized objects is also reported. This framework can also be used for the compression of images, where only an ASCII file describing the structural representation of the object is transmitted. The reconstruction of objects from their structural description is reported and improvements are suggested