Application Specific Standard Products with a 38.4GOPS Image Recognition Engine for Automotive Vision Applications

This paper describes Application Specific Standard Products (ASSPs), made up of an RISC processor, an image recognition engine, and peripherals, that have been developed as Systems on a Chip for automotive vision systems. These ASSPs are designed for automotive applications which use information about a vehicle's surroundings obtained from cameras and which need high speed processing for real-time systems control. A RISC processor and new image recognition engine are integrated onto a silicon chip in order to meet automotive specifications including cost and size. The image recognition engine based on local parallel architecture can compute a large amount of image data at high speed. The authors describe two ASSPs, one for general vision systems and the other for car navigation systems. They report on three practical applications to demonstrate the devices’ computational capabilities. These applications include road sign recognition for precise positioning; multitype lane marker recognition; and vehicle detection by voting edge-pair feature space. Sign recognition with one of the ASSPs, called SH7774, was especially successful in executing practical car navigation software and real-time sign recognition at the same time. The authors conclude that there are infinite possibilities for realizing safer and more easily operated vehicles by mounting cameras at the front, sides and rear to monitor and recognize the outside environment.