Parallel/Pipelined processor dedicated to visual recognition

A parallel/pipelined processor has been developed for visual object recognition. It detects the position and orientation of the top workpiece in case workpieces are overlapping. The processor consists of a real-time edge detector and a high-speed image correlator. The real-time edge detector detects contours of workpieces at the video rate of a solid-state TV camera. It is relatively insensitive to changes of ambient light through the use of a logarithmic Sobel operator instead of a standard Sobel operator. The high-speed image correlator detects the position and orientation of the top workpiece which has a complete contour without any hidden portion. Parallel and pipeline concepts were applied to make the image correlator carry out more than 100 million pixel-level operations per second.