基于轮廓特征理解的城市道路图像深度估计

Depth Estimation of Urban Road Image Based on Contour Understanding

准确估计道路场景图像中的深度信息,是智能交通和机器人导航中对障碍物估计和定位的关键。基于区域特征理解的单幅静态城市道路图像深度估计算法,可以通过边缘生长图像分割算法得到一系列封闭的图像区域;然后统计每个分割区域自身的多元特征,包括区域的颜色、面积、位置,所包含的直线、垂线和平行线;基于这些特征,进一步估计道路消失点,并实现天空、垂直面和道路区域的分割和三维空间推理,最后根据典型道路的深度变化规律实现对道路图像的深度估计。实验结果表明,该算法能够有效地估计道路消失点以及道路区域内部的渐变深度信息。

Estimating the depth of the road scene is a key step for depth estimation algorithm for single static road image is presented the image based on the edge growing segmentation algorithm intelligent transportation and robot navigation. A ~ The algorithm obtains a series of closed areas of The statistic characteristics of each region are computed, including color, area, position, straight lines, vertical lines, and parallel lines. Based on these characteristics, the road vanishing point is further estimated, and the sky, road and vertical regions are estimated. The depth of the road image is estimated by using vanishing point and the statistic characteristics. The experimental results show that the proposed algorithm can effectively estimate the road vanishing point, as well as the road depth information.