Efficient Vanishing Point Estimation for Unstructured Road Scenes

Vanishing point estimation is an essential and demanding task in vision-based road detection. One of the main limitations of the existing approaches for vanishing point estimation is computation efficiency, which hampers their real-time applications. This paper presents an efficient method for finding the vanishing point in unstructured road scenes. Color tensors are applied on the input image to find texture orientations and color edges. We propose new strategies to select optimized sets of vanishing point candidates and voters and to define the voting function. The proposed method is evaluated on a benchmark dataset of 2000 images of unmarked pedestrian lanes. The experimental results show that it achieves accuracy comparable with other state-of-the-art methods but with significantly reduced computation time.