VISION-BASED VEHICLE GUIDANCE

The use of microprocessors to control various automobile operations is now commonplace, and further computerization can be expected as resea rchers extend their efforts to develop autonomous, self-guided vehicle s. One of the most challenging research areas is road following, which requires the two basic functionalities of lane detection and obstacle detection. Thanks to the reduced costs of image acquisition devices a nd to the increasing computational power of current computer systems, computer vision has recently become a popular method for sensing the s urrounding environment. The authors use an approach that extracts and localizes features of interest, thereby limiting the computation-inten sive processing of images. A geometrical transform called inverse pers pective mapping makes a SIMD (single instruction, multiple data) appro ach practical for processing data captured in stereo images. Besides c ontributing to obstacle detection, the left stereo image is used in la ne detection. The use of a 3D surface called a horopter, moved onto th e road plane by electronic vergence, makes it possible to locate obsta cles and establish their distance and exact position in 3D world space . The authors describe the GOLD system, a stereo vision system develop ed at the University of Parma, Italy, for generic obstacle detection a nd lane localization. GOLD was first tested on an experimental land ve hicle for more than 3,000 kilometers along extra-urban roads and freew ays at speeds up to 80 kilometers per hour and is now being ported to the Argo autonomous passenger vehicle.