A Real-time Multi-lane Detection Approach Using Features Geometric Uncertainty and Stereo Vision

Mobile robots is getting popular with the public areas, such as airport, train station, etc. Among the technical methods for mobile robotic system, lane detection is an essential part, because lanes provide information for the robot system to keep current lane or switch to another lane space. However, existing lane detection algorithms are based on the assumption that the lane markings have good quality, but sometimes the lane markings are non-continuous and poor. In this paper, we propose a real-time multi-lane detection algorithm that is robust to the challenging lane condition. For basic lane detection part, we extract the lane features from images, and based on the features geometric uncertainty algorithm we can detect the lanes, guardrails, other vehicles in real-time. What' s more, the accuracy can be improved with the other vehicles direction information and guardrail depth information from stereo vision when the lane features are non-continuous or poor. Next, to prevent the false detection, an improved RANSAC (random sample consensus) algorithm is introduced in the algorithm. Finally, the proposed algorithm is evaluated on our dataset with different lane conditions. Experimental results show good robustness on variation of the lane conditions.