Road Information Extraction And Intelligent Warning Based On Monocular Vision

With the rapid growth of automobile usage and the development of the road traffic, the driver assistance has been an important research topic of Traffic Safety. The navigation and the intelligent warning of driver assistance systems are depended on the understanding of the road scene. Vision-based road information extraction technology is the key technology of road scene understanding. In this paper, the complex urban road and unstructured road information extraction and intelligent early warning method based on monocular vision technology have been studied extensively. The concrete research content includes the following aspects:(1) Firstly, in view of the complex urban road and unstructured road, road vanishing point detection algorithm based on the parallel and vertical envelope and the vanishing point detection algorithm based on dark color were proposed. The proposed algorithms use parallel lines and the linear of contours that segmented based on dark channels prior respectively. Then the vertical envelope lines are extracted to estimate the range of the road region. Next, a grouping strategy is proposed to compute the intersection of any two lines. Finally, the vanishing point locations are estimated by a clustering method. Combined the above two algorithms, a fast vanishing point detection algorithm also proposed.(2) Then, we proposed an algorithm that extracts the road region and the navigation information based on the vanishing point. It laid the foundation for the lane departure warning. Firstly, the road region is detected based on the vanishing point and the proposed soft voting border. Secondly, the navigation information is estimated based on re-using characteristics of cars driving in the road area extraction. Thirdly, the yaw angle and the warning time of the vehicle is calculated to obtain the yaw direction and decision-making method of the lane departure warning. This paper also further integrates the principles of dark channels to achieve a navigation line extraction using in the farm machinery.(3) Finally, a depth estimation method is proposed according to the vanishing point and recognition road region. The proposed algrithom obtained a series of enclosed regions using image segmentation algorithm. Then, each divided region’s characteristics and the road vanishing point are estimated. Based on these characteristics, the image is segmented into sky, vertical region and road region. At last, the depth of the image is estimated according to the depth variation rule of typical road. This study provides an important basis for depth information accurately predict and perception in Intelligent Transportation road navigation, obstacle and pedestrian detection.