Robust Lane Recognition Based on Arc Splines

SCHMİDT K. V. , Yeniaydın Y.

Digital Transformation Smart Systems, 24 - 26 Ekim 2018


This paper develops a general lane detection method and proposes new techniques for feature extraction and lane modeling. The proposed method first determines a static region of interest. Then, feature extraction is used to establish candidate lane pixels in a binary image. Next, the binary image is transformed to a bird’s eye view (BEV) via inverse perspective mapping. After that, a reliable region for the detection of the left or right lane markings is chosen based on the distribution of the candidate lane pixels on the BEV. Finally, a lane model is fitted to the extracted lane pixels. The paper further proposes a new Neighborhood AND operator for feature extraction and uses arc-splines as a lane model. In order to evaluate the quality of the proposed method, the paper performs an extensive comparison using different feature extraction methods and a second-order lane model. The experimental results show that the Neighborhood AND operator for feature extraction and arc-spline lane modeling are superior to the other techniques.