Abstract: With the rapid development of vehicle intelligence in these years, the automatic detection of road areas has been playing a more and more important role in the field. However, the detection of unstructured roads faces significant difficulties due to the fact that many unstructured roads do not have prominent lane marks or boundaries. Conducting the detection while setting the vanishing point as the constraint can substantially improve the performance of unstructured road detection. But in practical application, the existing methods for vanishing point detection of unstructured road generally have significant shortcomings of high computation cost and poor real-time performance. In order to improve the efficiency of calculation, a new method based on local directional pattern (LDP) texture feature for vanishing point detection of unstructured roads is proposed. Through the calculation of LDP texture features, the Kirsch mask is used to obtain the four-direction response amplitude of the pixel in picture, and the error of detection is reduced by the amplitude correction; and then the main direction of texture is obtained by calculating the corrected response amplitude; the vanishing points are selected by using the local adaptive soft voting method. Finally, the detection for vanishing points is achieved. According to the experimental results, the method is faster than existing methods and it can detect the vanishing points of unstructured roads accurately and effectively in the natural environment.

Key words: local directional pattern (LDP), vanishing point detection, unstructured road, local adaptive soft voting