Abstract
Cracks in 3D pavement data often show poor continuity, low contrast and different depths, which bring great challenges to related application. Recently, crack attributes, e.g. depth and width have attracted attention of highway agencies for maintenance decision-makings, but few studies have been conducted on crack attributes. This paper presents object-based image analysis (OBIA) method for crack detection and attribute extraction from laser-scanning 3D profile data with elevation accuracy about 0.25 mm. Firstly, a high-pass filter designed for pavement components in our previous research was applied to remove the fluctuation posture in 3D data, and then the smallest of-constant false-alarm rate algorithm was used to acquire lower point sets, including crack seeds and lower textures. Secondly, the objects were represented by above obtained 3D point sets and OBIA, especially, the depth statistics, shape and topological features of objects were described. Moreover, to enhance crack objects and remove texture objects gradually, multi-scale object selections and merges were conducted according to the local statistical characteristics differences of objects. Thirdly, the objects' orientation attributes were combined with tensor voting to connect and infer final crack objects, and then the object-level crack depth attributes could be extracted. The experimental results demonstrated that proposed method achieved average buffered Hausdorff scores of 94.39, Recall of 0.92 and F -value of 0.91 for crack detection on 30 real measured 3D asphalt pavement data. Furthermore, crack depth attributes can be extracted at different scales according requirements, the obtained location and depth attributes provide more comprehensive information for pavement maintenances.
Highlights
Cracks are one of most common and concerned diseases in pavement, to detect crack rapidly and accurately is of great significance for pavement maintenance and road traffic safety [1]–[3]
The generation of cracks can be considered as the gradual formation by local statistical characteristics differences between crack objects and surrounding texture objects in 3D data
A high-pass filter with frequency characteristics of pavement components was applied to remove the fluctuation of vehicle posture in 3D data, and the V-shaped structures in remaining data was acquired by constant false-alarm rate (CFAR) algorithm to form 3D point sets including crack region of interest (ROI) and lower textures
Summary
Cracks are one of most common and concerned diseases in pavement, to detect crack rapidly and accurately is of great significance for pavement maintenance and road traffic safety [1]–[3]. The elevations of cracks are usually lower than the normal pavement surface, so the high-precision laser-scanning 3D profile system can provide an effective data basis for obtaining pavement crack information [9], [10]. Crack detection technology based on laser-scanning 3D profile data is one of the hotspots of relevant pavement research institutions at home and abroad. Crack attribute information has gradually attracted attention of highway agencies for disease model analysis and decision-makings in recently years. Most existing methods focus on crack detection and classification, while few public studies mention crack attribute information, such as depth [6]. It is difficult to obtain suitable crack depth evaluation scale, the pixel level is accurate but with no practical significance, and the whole crack level is lack of local information
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
