Automatic recognition of surface cracks in bridges based on 2D-APES and mobile machine vision

Abstract

Compared with the artificial crack detection method, the bridge crack recognition method based on computer vision has the advantages of high efficiency, easy operation and low cost. However, under the condition of moving (UAV) shooting, the crack images collected often have quality defects such as low definition, complex background, severe interference by light and noise. Especially when faced with small cracks in early development, some traditional crack detection algorithms with high requirements on crack images cannot be well adapted. In this paper, an automatic recognition technology for surface cracks of bridges is proposed, which is suitable for mobile machine vision detection. The core of the technology is to obtain high-precision two-dimensional spectrum estimation of crack images by using two-dimensional amplitude and phase estimation method (ab. 2D-APES), and then to enhance the crack information by filtering low-frequency information, so as to realize the automatic recognition of crack targets in images. An industrial-grade drone (DJI Jingwei M200V2) equipped with a high-definition zoom image acquisition system was used to acquire images of the bottom and sides of the bridge of the Minpu Bridge in Shanghai. After locating, magnifying and cutting the apparent crack image of concrete, and then using the above method, the crack automatic identification was realized. Results show that the high-precision non-parametric amplitude spectrum analysis method can adapt to the situation of poor image quality of the UAV, and thus provides a feasible solution for the automatic identification of concrete cracks based on mobile machine vision.