Unstructured road detection from a single image

Image based road detection is a vital task for many real-world applications such as autonomous driving and obstacle detection. The detection of unstructured roads is particularly challenging by reason of blurry road borders, rough road surfaces and varying imaging conditions. In this paper, we propose a simple and effective method for unstructured road detection. The key idea is to consider the whole image as horizontal lines at regular interval. Based on the assumption that lower part of the road image contains much more road areas than other parts, we estimate horizon line by analyzing similarities between the very bottom line and other lines in terms of normalized cross correlation (NCC). A histogram based method is then carried out in the part below horizon line in order to predict road region. Finally, every point of line segments is classified as road or non-road by comparing the color feature with predicted road region. Further connection and dilation are conducted to extract road region from image. Experimental results on several complicated road images, both quantitatively and qualitatively, demonstrate the effectiveness and accuracy of our proposed method.