Segmentation Based on the Unstructured Road with Shadow

The road detection has a very important position in automatic driving and machine vision. This paper presents a method of unstructured road images with shadow, using two-dimensional Tsallis cross entropy and Hough line detection. First we convert the RGB image to HSI (hue, saturation, intensity) space model. Then we try to find the horizontal line of the vanishing point, with calculating the normalized cross correlation (NCC) values in intensity component. Using two-dimensional Tsallis cross entropy in hue and saturation components for single threshold segmentation, we subdivide the image into road region and the background region. After that, using the logic operations to them reduces the dimension. Finally, we use the Hough line detection to detect the road edge and predict the vanishing point. The experimental results demonstrate the effectiveness of our method.