Offset Vertical Stereo System for Real-Time Range-Finding to Preceding Vehicles

positions of each vertical projections of long horizontal We have developed a new real-time range-finding edges as particular characteristics of vehicles in front. system based on vertical stereo vision for measuring the distance to vehicles in front. In the system, stereo images are inputted from vertical cameras which are placed in the periphery of the windshield. However, the placement causes to the difference of vehicle's size in upper and lower images and missing stereo matching. To solve the problem, we proposed the powerful stereo matching algorithm that increases the missing. The method has two processes; the first process is to select particular characteristics of vehicles as matching candidates. The second process is to correspondence the same characteristic of the vehicles in the candidates using size invariant measure. To evaluate the accuracy of our algorithm, we compared measured values with actual distance. The result that the measured value is close to the actual one is indicated to few missing matching. Moreover, we implemented the algorithm on our high-speed generalpurpose video image processing system, ISHTAR, and processed video data in expressway scenes. We confirmed robustness of the algorithm under environmental changes, such as various vehicles in front, sunlight changes and shadows.