Research On Key Technologies Of Lanedeparture Warning System Under Monocular Vision

Lane Departure Warning System(LDWS) is used to remind drivers to take appropriate actions when the vehicle will deviate from the lane, in order to avoid the lane departure accidents and improve the security of driving. This paper mainly focuses on the key technologies of LDWS under the monocular vision, including calibration of vehicle-mounted camera parameters, edge enhancement of lane markings, identification and tracking technology of lane markings and lane departure warning decision. The study concretely lies in:1, Calibration technology of vehicle-mounted camera parameters. For internal parameters, Zhang’s calibration method based on AFT-CT430 high-precision calibration plate is adopted for offline calibration by the relation between pixel coordinate system and world coordinate system. The method is simple and of high precision. For external parameters, a three lines calibration method is used for online calibration according to the relationship of the slopes of three lines which are parallel to the vehicle axis. It does not require a dedicated site and it is easy to implement. Inverse perspective transformation is employed to verify the accuracy of calibration results, which provides good basis for determining region of interest(ROI) of road image and calculating the transverse offset of lane markings apart from the central axis afterwards.2, Edge enhancement technology of lane markings. Perspective transformation technology is used to identify the ROI of road image, and the weighted average method and improved median filter are employed to get the denoised grayscale image of ROI. Meanwhile, an improved edge detection method based on universal gravity is put forward to enhance edges of lane markings adaptively by the variable factor, which effectively realizes edge extraction of lane markings under the conditions of normal light, night and a large area of shadows.3, Identification and tracking technology of lane markings. Firstly, according to features of lane markings, line-parabola combination model is established. Then, Radon transform method based on constraint for conditions is proposed, combined with Kalman forecast tracking technology, to achieve the accurate recognition and tracking for line segment of lane markings. Eventually, a strategy of points scanning is used to collect edge point set of lane markings, along with least square method to fit the point set in order to get parameters of parabola segment of lane markings.4, Lane departure warning decision technology. For the disadvantages of existing models, a departure warning decision model based on transverse offset differential is designed, through combining the distance information between the vehicle and the lane markings with time information of upcoming lane departure, to achieve efficient lane departure warning. Experimental results show that the model has higher accuracy.