LSD-based adaptive lane detection and tracking for ADAS in structured road environment

Abstract

Lane recognition is important for safe driving in structured road environment; it is becoming an indispensable part of the advanced driver-assistance system (ADAS) for active security control. This paper proposes a novel lane detection and tracking approach for ADAS by using the line segment detector (LSD), adaptive angle filter and dual Kalman filter. In the lane detection process, the region of interest (ROI) within the inputted image is transformed into grayscale image, which is further preprocessed with median filtering, histogram equalization, image thresholding and perspective mapping. Then, the fast and robust LSD algorithm is applied on the ROI with the proposed adaptive angle filter to eliminate incorrect line segments more efficiently. In addition, a new three-level classifier based on the color, length and quantity of detected line segments is designed for lane classification. Finally, two Kalman filters are applied to track the detected lane and predict the following ROIs to improve the robustness and processing speed. The experimental results on four datasets show that the proposed method has robust performance in complex environments with the presence of shadows or other artifacts. It has average correct rate of lane detection higher than 94%, while the processing time is reduced by about 50% compared with a state-of-the-art method, and the average success rate of lane classification is above 85%.