Road Extraction At Bending Section Of Expressway

On expressway, traffic accidents happen mostly at the bending section. To reduce the number of accidents, intelligent video surveillance system for expressway has been widely used, with the bending section being the focus of monitoring. As an important step of the pre-processing of the system, road ROI extraction removes redundant information such as the sky and leaves, which narrows down the computation cost. And, since ROI is the base of the following work such as vehicle detection, behavior analysis, its accuracy ensures the good performance of the following functions. Additionally, as the conditions of the bending section are complex, the ROI extraction suffers large deviations. So, there is a strong need to do deep research on road ROI extraction on bending expressway, which is significant in practical applications.In this paper we focus on the ROI extraction of bending road, analyze the limitation of existing algorithms and the difficulties on this problem. A lane feature recognition algorithm is optimized. A vanishing point detection algorithm for bending road based on middle curve fitting is gave out, and A road edge curve fitting algorithm based on sub-regional line segments detection and clustering is proposed.As for the lane recognition, the reason of traditional algorithms’ poor noise immunity is analyzed. These algorithms include threshold segmentation and edge detection. Then the horizontal filter is optimized, detecting lane by extracting those pixels whose value reflect a pulse around its neighborhood in the direction slightly deviate from horizon.As for the vanishing point detection of bending road, an algorithm based on middle curve fitting is gave out. Firstly, an original vanishing point is calculated to generate the middle curve. Then the final vanishing point is determined by adjust the original one to fit the middle curve function. As for the road edge curve fitting, the road curvature’s smallness is utilized, we proposes an algorithm based on sub-regional line segments detection and clustering. By recognizing and clustering the edge segments, the control points are located, to accomplish curve fitting.This paper’s algorithms have been applied to the Huning Expressway Intelligent Surveillance System, and perform well.The innovations are as follows:· The horizontal filter for lane recognition algorithm. is optimized, which performs well in detecting lanes.· A vanishing point detection algorithm for bending road is gave out, which improves the accuracy of vanishing point detection in bending road scene.· A road edge curve fitting algorithm based on sub-regional line segments detection and clustering is proposed, which generates accurate ROI.