초록 |
This paper describes a stereo vision based obstacle detection algorithm, which is the core component of advanced driver assistance system incorporating lane departure warning, forward collision warning and avoidance. The proposed vision system recognizes the road lane, on which host vehicle is traveling, by template matching on the bird eye view of forward scene. The recongnition of road lane uses an assumption that a lane marking is a pair of neighboring rising and falling edge and a road lane is a pair of lane marking with a fixed distance. ROI(Region of Interest) is established according to the recognized ego-lane because preceding vehicle on the ego-lane is expected to be a potential threat to host vehicle. After the establishement of ROI, vision system generates disparity histogram by feature based stereo matching. Because the preceding vehicle has a large amount of vertical edges with the same disparity, it forms a peak in the disparity histogram. Consequently, the preceding vehicle can be detectable by simple threholding. The threshold of peak detection is designed to vary with respect to disparity, i.e.distance. considering the fact that obstacle appears smaller as its distance becomes further. Detected peaks are verified by the comparison of edge and color between left and right image. Ego-lane based ROI establishment and feature based stereo matching drastically reduce computational burden. Furthermore, disparity histogram based obstacle detection is proved to be robust because it captures big picture successfully ignoring the details. The effect of ego-lane based ROI and adaptive thersholding is verified by experiments with real vehicle. |