为了使道路场景的消失点检测能够适应不同的道路条件,提出基于全局图像特征的层次化消失点检测方法。通过全局图像特征提取全局道路特征,将道路分为4类并粗粒度定位道路区域。根据分类选择提取道路标识或边缘特征进行尺度变换的线段检测或区域分割并投票消失点集,再选择使用逆透视仿射变换或色彩纹理信息验证获得有效消失点。通过图像预处理移除道路车辆及阴影干扰,进一步提高检测精度。实验证明道路特征分类有效,在光照阴影、色彩纹理及遮挡等条件各异的场景中,层次化消失点检测方法均获得实时鲁棒的检测结果,比现有在复杂场景平均误差较小的基于本征直线方向与色彩纹理的检测方法精度与效率分别提高37.5%和20%。

To deal with different road conditions and environment situations, a hierarchical vanishing point detection method based on holistic image features is proposed. Four road types are classified based on an abstracted holistic road feature, which also coarsely locates the road. After classification, the lane or edges in a specific scale is chosen for the line-segment-detection or region segmentation, and for the voting of the intersection point. Additionally, an inverse projective image or region entropy verifies the result. Interference of vehicles and shadows are eliminated during image pre-processing to further improved the accuracy. Our results show that the hierarchical method can choose the right road type and robustly detect the vanishing point under complex road scenes in real time, achieving an improvement for accuracy and efficiency at 37.5% and 20% compared to the state-of-the-art method based on intrinsic line orientation and color texture properties.