Study On Adaptive Region Of Interest Based Lane-detection Algorithm

With the development of vehicular industry,and the increasing nation-wide car ownership,it is widely concerned to vehicular security for the sake of increasing traffic jams.In recent years,Advanced Driver Assistance Systems(ADAS)have a fast development,which is able to offer an excellent solution to assist vehicular-safety driving.as an important part in the ADAS,it is one of the key technologies to improve vehicular-safety driving situationLane-markings detection system contains two part content of lane-markings preextraction and lane-markings detection,of which this paper is to give a detail description:Lane-markings preextraction.firstly,the SROI(Static Region of Interest)is determined,in which the median filter is used to remove noise;secondly,the paper proposes a new lane-marking edge points segmentation method based on image gradient orientation and intensity on average.the ROI and the feather of lane-line gradient based lane-markings edge points are acquired,and then the noise is filtered out by the restriction of the lane edge points with gradient orientation.The lane-edge information is achieved,by which the gray-scale level standard deviation is redefined as the threshold and the gradient intensity on average is considered as the indirect segment object;finally the initial road vanishing point is located by using Hough transformation combined with Least Square method;Lane-marking detection.firstly,the vanishing point at the t frame acquired by using Lane-markings preextraction is used to redefine the DROI(Dynamic Region of Interest);secondly,the t frame vanishing point based line samples is structured,by which the lane-line parameters is fitted by using IRANSAC(Improved Random Sample Consensus)algorithm,and the t+1 frame vanishing point and t+2 frame DROI are certificated.Meanwhile improved Hough transformation and improved IPM(Inverse Perspective Mapping)based lane-line fitting comparative experiments are conducted.thirdly the curves-line is rebuilt by using hyperbolic model;finally this paper proposes a method to achieve lane dynamic tracking,which is according to the DROI and the characteristics of inter-frame lane information changing a litter.The test results indicate that it is efficient to hold back the noise and extract the lane information under complex condition by using pixel gradient based image segmentation approach;at the same time,it has a good performance in lane detection with the method of IRANSAC algorithm under complex situation,such as rainy day,night,shadow and broken lane markings.Meanwhile hyperbolic model is used to reconstruct the curves,which ensure the less time consume and accuracy of lane detection;finally it is able to achieve lane tracking and has a high-accuracy lane detection under lane changing,turning and lane marking disappearing condition.