Research On Road Detection Algorithm In Vehicle Vision Navigation

After the road environment information is obtained from the vision sensor in the vehiclenavigation system, the system can analysize and process it, ultimately making intelligentdecisions, assisting the driver to control the vehicle, accidents can be avoided to some extent,ensuring the safety of vehicles and pedestrians.Roads can be divided into two types of structured and unstructured. When the vehiclenavigation uses visual techniques, for the structured road, this paper makes a specific detectionmethod. Because the lane lines are the most obvious characteristics of road, this paperimplements the detection of road by extracting the lanes. In visual navigation, road detectingrequirements of a good real-time performance, high positioning accuracy is demanded on theroad. Basing on the previous detecting methods, this article makes some improvements, whichcan lay the foundation for further study of automotive driver assistance systems.For the straight lane detection, first, the image collected complete preprocessing,including graying, median filtering, edge detection and binarization. This paper in the case ofdetails reserving without loss of useful information uses the weighted average method to graythe image; the median filter method is applied to filter out the noise; the edge detection isapplied to find the lane edge; Otsu threshold method is applied to binarize the image. Then, theregion of interest is determined in the image after the preprocessing; Probabilistic HoughTransform is used to detect the straight lines; Least Squares method is used to fit the lines inthe original image.For the curved lane detection, the image will be divided into two parts, both near and far.Application of piecewise linear method, to some extent, improves the applicability of thealgorithm. First, pretreated, the regions of interest of each part are determined, the lines aredetected and fitted. Then, the fitted parameters are applied to calculate the intersection of linesof each part. The bending direction of the road is determined by the slopes of lines and therelation of two projections of intersection points on the horizontal axis.For the road tracking, Kalman filter is applied to track the lane lines. The lane lines aredetected in the initial frame, the information is recorded, its position of next frame ispredictied. The system is constantly updated, determining the dynamic region of interest. Then, the lane lines are detected and fitted. The intersection of two lane lines and their intersectionswith the region of interest are finded. The lane lines could be fitted in the original image bythese parameters. The algorithm using tracking technology improves the accuracy of detection.For lane departure warning, a combination of image information, road information andcamera parameters CCP method is used. The geometrical quantity of the left and right roadline, camera parameters and standard quantitative are used to calculate the current parametersof the vehicle, in order to achieve lane departure warning.