Research On The Lane Detection And Classification Of Structural Road

In recent years,in order to reduce the frequency of traffic accidents,research on intelligent vehicle safety driving assitant system based on visual has become a hot spot for that lane recognition is the indispensable important part and significant factor to improve the driving satety.In this paper,an algorithm for real-time detection of structured road lane and its classification is proposed.The first is image preprocessing given based on the dark channel prior to remove haze of road image,determine the region of interest for subsequent image processing,image color space conversion to YCbCr color space,use median filter method to filter image.Secondly,aimed at the light effect on the image quality problem,this paper proposes a suitable image binarization method for detection of white and yellow lane.The method is based on white in the Y component of YCbCr color space and yellow in the Cb component has an invariant property,so it has strong robustness to many kinds of illuminant conditions and low complexity of computing.In order to accurately detect the lane line,this paper selects the candidate lane in the binary image partition and fit the left and right lanes with an improved partitioned probabilistic hough transform algorithm based on the dynamic setting of the accumulator threshold.Then extract taxonomic features through the analysis of the different characteristics of the common structured road lane to classify the lane line,which is divided into ten types of white or yellow single line(dashed line,solid line),white or yellow double line(solid-dashed line,dashed-solid line,double solid line) and in the classification process using the naive bayesian classification algorithm to improve the reliability of the classification.Besides,due to the lane recognition is generally used for lane departure warning system and lane change assistance system,real-time processing vehicle video image is required.Therefore,this paper puts forward a lane tracking detection algorithm to further improve the system's real-time and robustness,which uses Kalman filter to dynamically determine the region of interest. When the vehicle is running more stable and the lane recognition is better,the lane tracking detection algorithm is called;otherwise,the initial lane detection algorithm is called.Finally,the experiment on 1560 frames of the road image show that the proposed algorithm can work well in a variety of illuminant and has satisfactory performance,with a certain real-time,robustness and reliability.