Detection And Recognition Of Novel Lanes Based On Fisheye Camera

In modern society, the using of cars has generally brought a lot of problems such as traffic accidents, environmental pollution and so on. Among them the incidents of auto deviating from the driveway account for a large percentage of traffic accidents. Therefore, the research on Lane Departure Warning System has drawn more and more attention from every country in the world. The lane detection and recognition is the basic and essential function. Up to now most of lane detection and recognition methods have been proposed. However, not a method has been found so far to efficiently identify the lane which is composed of raised pavement markers.Fisheye camera is a mean of acquiring external environmental information and an algorithm is designed to identify the novel lanes. In order to identify the lane which consists of raised pavement markers, two kinds of methods are applied in this paper:One is drawing lessons from a filter for detecting lung nodes in medical images we design a dot filter for detecting the bright dots which is suspected raised pavement markers. Moreover, the partition of regions of interest makes us focus on the objects on the road, which can reduce the interference of the bright dots that is not raised pavement markers. Another is to use OLS template for matching. According to the position of a raised pavement marker on the road and its real size, inverse perspective mapping can be applied to obtain its shape on the image and the OLS template is constructed for template matching. In this way we are supposed to get an accurate position of a raised pavement marker and get rid of the existing noise points after using filter. A combination of these two methods is able to efficiently detect the raised pavement markers on the road. After getting the center points of raised pavement markers a novel algorithm based on Hough transform and clustering is proposed in this paper to fit the lanes. The algorithm can be used when there are only a few of center points of raised pavement markers and it has noise immunity for the noise points got by error detection.We make experiment and evaluate the algorithm in scenarios with different kinds of lanes. The results of experiment and evaluating demonstrate that the method for detecting raised pavement markers and the lane recognition algorithm are robust and efficient.