Automatic detection, localization and quantification of structural cracks combining computer vision and crowd sensing technologies

Abstract

The emerging of computer vision based methods provides the capacity to detect and quantify the structural damages efficiently. Moreover, the integration of high-performance sensors in the mobile phone allows for the efficient acquisition of multi-source heterogeneous data of damages such as images, locations, etc. This study proposed a computer vision and crowd sensing based method to detect and quantify structural cracks for the distributed in-service bridges. A previously established crowd sensing system was employed to acquire bridge structural cracks and the corresponding locations, which is consisted of a mobile application (APP) and the cloud management platform. As many as 15 volunteers were mobilized to use the APP to collect the multi-source heterogeneous data of cracks from in-service bridges, which consists of 223 crack images, covering 5 typical bridge structural components, i.e., the guardrails, the road surfaces, the beam undersides, the abutments and the piers. Then, the classic U-Net model was trained to segment the crack regions from the image background. Finally, an image skeleton-based processing method was proposed to acquire the pixel size of the cracks for quantitative evaluation. Testing results show that for images with a resolution of 256 × 256 pixels, the crack width is within 10 pixels, while the crack length ranges from 200 to 500 pixels, the error compared to the actual value is within 5 %. Crack images from different scenarios were used to test the applicability of the proposed method. This study provides a novel method to alleviate the demand for professional inspection engineers and vehicles, and validates the feasibility of the proposed method in practical application.