Research On Lane Detection Method Based On B-Spline Curve Model

Lane detection is the basic task of lane deviation warning and lane maintenance in advanced assisted driving.At present,lane detection methods based on traditional image processing are quite common,which mainly include road image preprocessing,lane detection and lane tracking.Road image preprocessing is to first set the ROI according to prior experience,and then conduct grayscale,binarization and lane edge feature extraction within the ROI.Lane detection is usually to select an appropriate road model,process the edge points to obtain the target points and conduct fitting.Normally,lane models are linear model,parabola model,B-spline curve model and combination model.The purpose of lane tracking is to predict the position of the next frame of lane according to the correlation between adjacent two frames of images.Lane tracking is realized in a local area,so the real-time performance is high.If several consecutive frames of road images are not tracked to the lane,the lane line detection will be conducted again in the global scope.Kalman filter is a common tracking algorithm.The above method has several problems.The common edge feature method does not have a strong filtering effect on the environmental interference,and the acquired edge feature image has a lot of interference or serious feature loss of lane line.At the same time,common road models have limitations and are not suitable for all kinds of common roads.In addition,the poor fitting performance of the common lane detection to the edge key points is easy to cause the lane mischeck and missing check.The lane detection method based on deep learning has high detection accuracy,strong learning ability,wide application range,data-driven and good portability,but there are problems such as model training can only be carried out offline,large amount of calculation,high hardware requirements and complex model design.In view of the problems existing in the above methods,this paper firstly uses Canny operator to detect the edge of road images,and carries out morphological denoising processing on the obtained edge images,so as to accurately obtain the lane edge images.Secondly,the image segmentation method and dynamic programming algorithm are used to obtain the candidate set of control points,and the lane is detected by combining the random sampling consistency algorithm and cubic B-spline curve model.Thirdly,aiming at the problems of missed and misdetected lanes,the kalman filter method is adopted to establish the lane parameter tracking model,construct the calculation equation of process noise and observation noise,and track the coordinates of four control points of B-spline curve and the running state parameters of vehicles.Finally,after a lot of experimental analysis,the lane detection method in this paper has good instantaneity,accuracy and robustness.