基于多特征检验的三维沥青路面裂缝检测

针对由裂缝对比度低、路面纹理复杂多变等因素引起的沥青路面三维图像的裂缝检测精度低的问题，对原始三维裂缝图像进行尺寸降维、灰度校正、高斯滤波等预处理；然后以图像截面为研究对象，分别对4个方向的截面依次进行特别设计的倾斜度、高斯分布、边缘梯度3种特征检验，从而获得裂缝截面；接着对各个方向的裂缝截面进行融合和去噪，获得完整的裂缝二值图像；最后，根据路面粗糙度的高低，变化高斯分布特征检验中的相关参数，实现裂缝的高精度检测. 研究结果表明:提出的算法能达到89.19%的准确率、93.69%的召回率及91.06%的 F 值，优于基于三维光影、种子识别的典型三维图像裂缝检测方法. 

Crack Detection of 3D Asphalt Pavement Based on Multi-feature Test

In order to solve the accuracy problems in the crack detection of 3D asphalt pavement, which are mainly caused by low contrast between cracks and the surrounding area and complex pavement textures, a three-step preprocessing was conducted on original 3D images firstly, including size reducing, intensity correction and Gaussian smoothing. Then, three predominant feature tests of tilt-level, Gaussian-distribution and edge-gradient were applied to the image profiles of four directions successively so as to obtain the crack profiles. Moreover, the crack profiles of four directions were merged and denoised to acquire the intact cracks. Finally, according to the roughness of pavement surface, a related parameter in the Gaussian-distribution test was adjusted to realize the crack detection of high accuracy. The experiment result indicates that the proposed algorithm can reach 89.19% of accuracy, 93.69% of recall and 91.06% of F-measure, which outperforms another two typical 3D recognition algorithms based on the theories of 3D shadowing and crack seeds.