No abstract

This paper presents a real-time vision system as a tool of developing robust vision algorithms. Our goal is to develop robust vision algorithms for complex scenes, such as city streets, and for various outside conditions, such as weather, lighting and shading. Since no ordinary computer has enough computational ability for developing vision algorithms efficiently, the system has been developed in the form of compact image processing hardware. We have developed special hardware modules for extracting image features, namely edge segments, depth from stereoscopic and optical flow. A high speed template matching hardware module has been also developed for detecting and tracking objects. Each hardware consists of one or two 6U format VMEbus compatible boards and operates at a video rate due to the pipeline architecture. Experiments have been done on this vision system as follows: road following from edge segments, moving objects detection from optical flow on the images of moving camera, obstacle detection from depth map based on stereoscopic and tracking preceding vehicles by a template matching method. The results obtained are found to be encouraging, indicating that the vision system is useful for developing vision algorithms and that the algorithms based on a part of the proposed ideas are robust for recognition of complex road environment.

No abstract

No abstract

No abstract