A Visual Brain Chip Based on Selective Attention for Robot Vision Application

摘要

Due to huge computational loads and the state of art of current hardware, nowadays how to implement real-time vision tasks on a signal chip is still a great interesting work. It is critical for eventually accomplishing biointelligence on robots, whatever humanoid robots, insect robots or even space robots. This paper presents a design of a so-called visual brain chip for robotic vision applications in order to explore biological visual information processing mechanisms. Biological vision systems explore their environments via allocating their visual resources to only the interesting parts of a scene. This is achieved by a selective visual attention mechanism that controls eye movements. Such mechanism is useful to reduce the complexity of calculation. The computational process was divided into three stages for achieving real-time visual computation. The whole system for the chip design used a PTZ TV camera, ASIC technology altogether with a LEON RISC processor and was verified on the FPGA development board. It is a good attempt for future robot vision system applications, especially for space robot, due to limited onboard computing resources, limited viewpoints and mobility.