Lane Detection for Track-following Based on Histogram Statistics

Abstract

Visual navigation technologies such as lane detection have been applied in many fields. A multi-line detection algorithm based on histogram statistics is proposed for the track-following application. After the preprocessing of the original image and projecting, the pixel histogram in the bird view space can be got and the starting points of the lane detection are obtained by filtering and clustering the histogram. Subsequently, the sliding windows are moved to capture the pixels on the lines. Finally, the quadratic curves are fitted as the model of the lines and are projected back to the original image space. Compared with the current other feature-based lane detection algorithms, our algorithm can deal with multi-curve or cross horizontal lines better in the track-following application with better robustness.