A vision system for detection and tracking of stop-lines
Published in IEEE-ITSC-14, 2014
This paper presents a computer vision algorithm that detects, by analyzing lane-marking detection results, stop- lines and tracks, using an unscented Kalman filter, the detected stop-line over time. To detect lateral and longitudinal lane- markings, our method applies a spatial filter emphasizing the intensity contrast between lane-marking pixels and their neigh- boring pixels. We then examine the detected lane-markings to identify perpendicular, geometry layouts between longitudinal and lateral lane-markings for stop-line detection. To provide re- liable stop-line recognition, we developed an unscented Kalman filter to track the detected stop-line over frames. Through the testings with real-world, busy urban street videos, our method demonstrated promising results, in terms of the accuracy of the initial detection accuracy and the reliability of the tracking.
Young-Woo Seo and Ragnunathan (Raj) Rajkumar, A vision system for detecting and tracking of stop-lines, In Proceedings of the 17th IEEE International Conference on Intelligent Transportation System (ITSC-14), pp. 1970-1975, 2014.