Lane Detection Based on Ternary Tree Traversal

In view of the complexity and inaccuracy of lane detection algorithm, a new method based on ternary tree traversal was proposed. Firstly, the binary image was obtained by image preprocessing, including the adaptive selection of region of interest(ROI), image segmentation and dilation. Secondly, candidate feature points indicating the center line of lane were extracted under the constraint of distance threshold. Thirdly, to obtain their position distribution information, the candidate feature points were traversed using the ternary tree traversal; a filtering algorithm was presented to filter the pseudo feature points in the candidate feature points and get effective feature points, which was designed according to the quantity constraint of consecutive candidate feature points and the parameters constraint of its fitting line. Finally, the effective feature points were fitted by RANSAC(random sample consensus), and then the equations of lanes were obtained. In addition, in order to improve the robustness of the algorithm, especially when there is no lane in current frame, a prediction model was put forward. Test results show that the proposed method can detect lane quickly and effectively in most of the complex road conditions, such as part of lanes missing, vehicles passing and symbols on the road surface existing, etc. For a color image of 640×480 pixels, the average cost time is about 116ms, and the detection rate reaches 91%.