Transputer-based 3D object recognition

Concerns the recognition of objects arriving on a conveyor belt, passing into the field of view of a fixed camera. The distance of the object from the camera is known, and the object is in one of a small number of stable positions. There remain three degrees of freedom: the object when viewed may be anywhere on the surface of the conveyor belt, and may be rotated at an arbitrary angle around a vertical axis. The computational task is to identify an image as representing one of a number of known objects, and to report the position and orientation of the source object. The transformation of an object can be found using the data from a vertex and an edge directional vector. An algorithm is developed. There are several stages in matching an image to an object. A set of identifiable points is needed. Vertices where edges join are suitable. Then, the pairing of vertices from image and object has to be decided. After this data from two vertex pairs is used to find the rotation and translation that is required, and finally the remaining vertex pairs are checked for a full set of matches. If the full match fails, then the process back-tracks. The problem is attacked with T414 transputer networks and 3L parallel C programs in order to provide a solution in the shortest time comparable with modern computing resources