A new automotive vision platform has been developed for practical applications. The vision platform simultaneously realizes high-performance computing power and reliability for automotive use by using a newly developed dedicated image processor. The developed processor has specialized and novel hardware allowing it to process a large amount of image data at high speed under a moderate clock frequency. The vision platform has enough capability to process multiple applications at the same time. In this paper, we describe the unified memory architecture in the vision platform. We introduce specific functional units for various applications such as the edge analysis unit. We report results from experiments with a real-time on-board lane recognition system.