Real-time lane markings recognition based on seed-fill algorithm

Road markings express the rules for the road while catching the upcoming road layout. These rules are applied to driving scenarios by real drivers who are known to Highway driving manuals. However, autonomous driving vehicles must read the roads in the same way that real drivers do. In this paper, the problem of automatically reading the road markings is addressed. A new approach for lane classification using the onboard camera is presented. As an initial step, the road boundaries are detected using the Hough transform model. The region of interest (ROI) is divided into two sub-regions. Hough Transform is applied to each of the sub-regions independently. The computational time required for lane detection is improved in this way. After that, adaptive smoothing and some processing steps are added to reduce the noise while still making the close edge's apart. Then, Seed fill algorithm is applied to the lanes location to identify the lane markings types. This method is able to recognize five types of lane markings such as: dashed, solid, double solid, dashed-solid and solid-dashed. The method is applied to a large set of video sequences with various situations and showed that the accuracy is over 95%.