Research On The Approach Of Structural Information Based Road Detection

Road detection is one of the important issues in visual navigation system of the UGV.Due to the fact that a vision sensor has the ability to capture more information and it is more cheaper than a laser-based sensor, vision-based road detection has been an active field of research for past several decades. However, vision-based road detection is still a challenging problem due to the diversity of road scene. This paper mainly focuses on vision-based road detection by integrating structural information into the road detection task. The main works in this paper are as follows:1. A robust vanishing point detection method is proposed for fast estimation of the road vanishing point in challenging scenarios. The method employs multi-population genetic algorithm(MPGA) to search vanishing point candidates heuristically, and the value of the MPGA fitness function is obtained by a novel voting scheme. The proposed vanishing point detection method is both efficient and effective in vanishing point detection;2. A noval seed selection method is proposed for superpixel-based GrowCut. Road vanishing point is used in this paper to define the road and the background clustering area,and then the GrowCut seeds are chosen automatically by using the K-means clustering algorithm. The proposed seed selection method is rubost in challenging road scene;3. Superpixel-based GrowCut algorithm is proposed for road detetion, and illumination invariant and color features are combined to measure the distance between two superpixels. Thus, the robustness of road detection is improved;4. A high level information based method is proposed to refine the road segment.Conditional random field(CRF) is used to integrate road vanishing point information and road shape prior information into the road segment, and to ensure that the labelings are globally consistent.To validate the proposed method, this paper performs several experiments to evaluate the average performance, scale invariance ability, noise sensitivity and fog sensitivity of the proposed method. The experimental results illustrate that the proposed method exhibits high robustness compared with the state-of-the-art.