基于深度图像的机场道面裂缝自动检测算法

为了实现强噪声、弱光照、低对比度条件下的机场道面细小裂缝检测, 设计了基于深度图像的机场道面裂缝检测算法; 将采集到的深度图像划分成多个网格, 并对每个网格进行扩充, 获得了局部道面区域; 针对每个网格区域, 基于随机抽样一致算法进行局部三次曲面构建和优化估计; 在此基础上, 在全局尺度下融合全部网格区域的曲面模型, 生成整个图像采集区域道面的全局曲面模型; 利用全局曲面模型与原始深度图像之间的差值图像, 采用自适应阈值方法分割出候选裂缝像素, 并利用裂缝的像素总数、长度以及长宽比等多种形态学约束筛选候选裂缝像素, 去除错误的候选裂缝像素, 从而获得了最终的裂缝检测结果; 在机场道面深度图像数据集上进行了试验, 以人工标注结果作为真实值, 以准确率、召回率以及F值作为量化评估指标, 将提出的算法分别与4种有代表性的传统算法进行了对比。试验结果表明: 传统算法能够取得的最高准确率、召回率以及F值分别为77.05%、41.02%和50.02%, 提出的算法在准确率、召回率和F值3个指标上均有明显优势, 其均值分别为91.20%、97.99%和94.12%;提出的算法能够在分辨率为1 984像素×2 000像素的深度图像上检测出最小宽度为3 mm、最小长度为10 cm的裂缝, 实现了在复杂机场道面场景中识别细小裂缝的目标。 

Automatic crack detection algorithm for airport pavement based on depth image

To detect small cracks in airport pavements under strong noise, weak illumination and low contrast, a crack detection algorithm for airport pavements based on depth images was designed. The collected depth image was divided into multiple grids, and each grid was expanded to obtain a local pavement region. For each grid region, the random sampling consensus algorithm was used to construct and optimally estimate the local cubic curved surface. On this basis, the global curved surface model of the entire image acquisition region of parement was generated by fusing the curved surface models of all grid regions under the global scale. Based on the difference image between the global curved surface model and the original depth image, candidate crack pixels were segmented by the adaptive threshold method, and various morphological constraints, such as the total number, length and length width ratio of crack pixels were used to screen the candidate crack pixels to eliminate the incorrect candidate crack pixels, so as to obtain the final crack detection results. The experiment was carried out on the airport pavement depth image datasets. The manual annotation results were taken as the ground truth. The accuracy, recall rate and F value were used as the quantitative evaluation indices. The proposed algorithm was compared with four representative traditional algorithms. Experiment result shows that the highest accuracy, recall rate and F value of the traditional algorithm are 77.05%, 41.02% and 50.02%, respectively. The proposed algorithm has obvious advantages in the accuracy, recall rate and F value, with average values of 91.20%, 97.99% and 94.12%, respectively. The proposed algorithm can detect the crack with the minimum width of 3 mm and the minimum length of 10 cm in the depth image with a resolution of 1 984×2 000, and realize the target of detecting small cracks in the complex airport pavement scene.