融合多特征与格式塔理论的路面裂缝检测

摘要

Pavement cracks are often mixed with random particle textures on road surface and a variety of interference under natural environment, results in the crack detection method based on single feature cannot recognize real crack accurately. Therefore, this paper presents a novel pavement crack detection method through integrating multi-features fusion and Gestalt principles. It extracts the intensity differences, the probability of occurrence and edge property of cracks in multi-scale local regions as low-level salient fea-tures firstly. Then, according to the texture inhomogeneity and the spatial continuity of the irregular curvi-linear structures of cracks, a novel texture anisotropy measure method (LFIA) is presented, which can weaken the disturbance of noisy points and pseudo-crack fragments efficiently. Based on the similarity, proximity and integrity principles of Gestalt theory, this paper adopts iterative clipping method to pre-segment LFIA map and proposes a crack spatial consistency enhancement strategy based on in-tra-regional and inter-regional connectivity to extract cracks. The experimental results of various collected pavement crack image database show the outstanding anti-noise performance and robustness. The precision and recall of our method is significantly superior to several existing conventional algorithms.%路面裂缝常常混杂着随机的路面颗粒纹理和自然环境下的多种干扰，基于单一特征的检测方法无法较为准确地提取裂缝，为此提出一种多特征融合与格式塔理论相结合的路面裂缝检测算法。将多尺度局部区域中裂缝的灰度差异、出现概率以及边缘特性作为低层显著特征，根据裂缝纹理的不均匀性，结合裂缝不规则曲线结构的空间延续性，提出一种新的纹理各向异性度量方式(LFIA)，以高效削弱噪声点与伪裂缝的干扰；然后引入格式塔理论中的相似性、接近性和完整性原则，采用迭代剪裁预分割LFIA图，基于区域内部以及区域间连接度的裂缝空间一致性增强策略，突出裂缝。在收集的各类裂缝图像数据库上的实验结果表明，该算法抗噪性能好、鲁棒性强；裂缝提取的准确性、完整性要优于已有的算法。