模糊航空图像中的道路自动检测方法

Automatic detection method of roads from fuzzy aerial images

为了在模糊航空图像中精确地检测道路,通过分析图像中道路特性,提出了一种道路自动检测方法。通过多尺度Retinex算法增强模糊图像,用改进的Canny边缘检测算法检测图像中的主要路段,使用交叉熵理论和贝叶斯决策理论自动获取梯度图像中的高低阈值,从而将灰度图像转化为二值图像,并将图像中所有线性目标进行骨架提取。根据线性目标的形状与尺寸参数进行噪声滤除,并根据端点的方向与端点间的距离进行道路间隙缝合,并结合边缘和原始图像信息调节和修正已检测出的道路。将道路自动检测方法与几种常用的图像分割算法进行比较,包括大津阈值分割算法,Canny边缘检测算法与图论最小割算法,并使用道路自动检测方法对模糊图像中的单条道路、交叉道路和多条道路进行检测。检测结果表明:对模糊或光照不均的航空道路图像,Retinex算法增强图像后可以清晰显示主干道路,而常规的图像分割算法无法将主干道提取出来,使用改进的Canny边缘检测算法并附以图像后处理功能较好地提取主干道路。使用道路自动检测方法能够清晰地检测模糊航空图像中单条道路、交叉道路和多条道路,与人工识别的效果接近。

In order to accurately detect the road from fuzzy aerial images, an automatic road detection method was proposed based on the characteristics of roads in images. The fuzzy images were enhanced by using multiple scale Retinex algorithm. The main road segments in images were detected by using improved Canny edge detection algorithm, and the high and low thresholds in gradient images were automatically obtained by using cross-entropy theory and Bayesian iudgment theory, the gray image was transformed into the binary images, and the skeletons of linear target in the image were extracted. The noise was filtered based on the shape and size characteristics in the linear target, the gaps between segments were linked based on the curvature and the distances between segments, and the detected road was adjusted and modified by combing the edge and the original image information. The proposed automatic road detection method was compared with several widely used traditional algorithms, such as Otsu threshold segmentation algorithm, Canny edge detection algorithm, and graph theory based on the minimum segmentation algorithm, a single road, cross roads and several roads in fuzzy images were detected by using the proposed road detection method. Detection result indicates that as for fuzzy or uneven-illumination aerial road images, the trunk roads can be clearly displayed after enhancing images by Retinex algorithm, while the conventional image segmentation algorithm can not do. The trunk road can be well extracted by using the improved Canny edge detection algorithm with image post-processing function. In the detection of single road, cross roads and several roads in the fuzzy aerial images, the target roads can be clearly detected by using the proposed method. The effect of detection method is close to the result of artificial recognition. 20 figs, 21 refs.