Research On Freeway Lane Detection For The Aerial Video

In recent years,the accident on the freeway is being more and more serious.The freeway monitoring method based on the unmanned aerial vehicle,can cover wider range,which makes it to be one of the most promising freeway monitoring methods and lane detection to be a research hot spot.After deeply studying the existing algorithm,a lane detection algorithm for aerial freeway video has been proposed.The main research contents are as follows.(1)Research on image segmentation algorithm:In order to detect the lane feature from aerial images which have complicated background,a segmentation method for aerial freeway images has been proposed.First of all,for reducing the processing time,a down-sampling method based on peak interpolation is presented to reduce the image resolution,while retaining the information of the lanes.Then,aiming at the problem that the aerial freeway image has complicated background,a lane enhancement method based on the enhancement of saturation image is presented,which increases the color difference between lanes and pavement.And next,a color image segmentation algorithm based on RGB color space is further presented.Results shows the lanes can be segmented from the aerial freeway images by the proposed algorithm accurately.(2)Research on clustering algorithm for lane feature points:In order to deal with the data of each lane separately,a clustering algorithm for feature points of the lanes has been proposed.First,in order to increase the similarity of feature points on the same lane,a similarity measurement method based on line spacing is proposed,solving the problem caused by the lane marking tiling and bending,dotted line and small distance between two lanes.And then the feature points are clustered based on the spectral clustering algorithm.Results show the feature points can be clustered by our proposed algorithm accurately.(3)Research on model selection and parameter estimation.In order to better describe the shape of the lanes,a cubic B spline curve is designed as the lane model,which can indicate the lane longer with higher accurate.In order to estimate the parameters of lane accurately,a parameter estimation algorithm based on RANSAC,lane distribution and Kalman has been proposed.Firstly,parameters of lanes are estimated based on RANSAC.And then parameters of the wrong lanes are removed according to the distribution characteristics of the lanes in aerial image and corrected based on the Kalman algorithm.Results show the parameters of lanes can be estimated by the proposed algorithm accurately.(4)Software design.In this paper,a software of the lane detection for aerial freeway video and vehicles' behavior monitoring based on the proposed algorithm has been finished.After carefully analyzing the requirements of the system,the architecture design,module design and programming are completed.The results show that the software can accurately detect lanes and vehicle violations.