Lane boundary detection using an adaptive randomized Hough transform

To detect lane boundaries robustly, R channel and B channel of color road image were used to form a gray level image. Size of the gray image was reduced and Sobel operator with very low threshold was used to produce gray edge image. In adaptive randomized Hough transform, pixels of gray edge image were sampled randomly according to their weights corresponding to their gradient magnitude. 3D parametric space of parabolic curve was reduced to 2D and two parameters were estimated by use of gradient direction, then another parameter was used to verify the estimated parameters by adaptive threshold value. Such lane markings can be detected accurately and robustly. Experimental results in different condition prove validity of the method.