Driver assistance systems using an optimized RANSAC B-spline fitting algorithm

Abstract

Driver assistance systems (DAS) are one of the most important technological means for modern vehicles to ensure driver safety and decrease vehicular accidents on roads. The systems include the localization estimation of the lanes and the determination of the relative position of vehicle and lanes. In this paper, we present a robust driver assistance systems designed to detect and track lane markers from image sequences captured by a front-view on-board monocular camera, and to raise an alarm when a lane departure occurs. The lane detection and tracking is based on inverse perspective mapping (IPM) for the region of interest (ROI) of a camera sequence, filtering by a selective oriented Gaussian filter, the Hough transform, and Kalman filter to give initial regions to our optimized random sample consensus (RANSAC) B-splines fitting. The proposed system works well on all kinds of roads, with curved or straight lane markers and in various weather conditions. Our experimental results and the accuracy evaluation indicate that the proposed lane detection and departure warning algorithms can run robustly in real time, achieving an average speed of 32.32 ms per frame for a 320 × 240 pixel image and 41.64 ms for a 640 × 480 pixel image, with a correct detection rate over 90.829%. © 2014 International Information Institute.