Lane recognition system based on the high-speed pattern matching method

Abstract

Applications of lane recognition using an on-board camera in the real world require image recognition technology that is both fast and robust. The authors propose a method that takes structure matching as a search problem for optimal patterns and then uses a voting approach for discrete template patterns as a method for pattern matching in this kind of image recognition technology. The authors implement their proposed method as an on-board device for lane recognition, then set it up in pipelined hardware, and confirm that it can work in real time. In addition, the results of comparing their method with a conventional lane recognition method using 1000 scenes of images of highways in the rain, a subject used for lane recognition, confirm that their method makes few errors and is robust. © 2004 Wiley Periodicals, Inc. Syst Comp Jpn, 35(8): 45–53, 2004; Published online in Wiley InterScience (). DOI 10.1002sscj.10591