Investigating the effectiveness of the sobel operator in the MCA-based automatic crack detection

Abstract

The life expansion of infrastructure systems is a common social issue in modernized countries and then early detection and treatment is the best solution on the issue. According to the shortage of the human resource for expert inspections, an automated detection of faulty points in the infrastructure is highly expected nowadays. Concrete crack detection in parts of bridges is a keen target in the robotic automatic inspection by using a drone with the high-resolution proximity camera. Taking fine pictures itself is not an issue if it is flying in the stable condition; however a micro vibration, intensity and/or contrast instability prevent a measurement under mm width of concrete cracks. Detecting the fine feature via images obtained from the drone attaching on the bridge deck requires an appropriate decomposition of different textures in image such as anisotropic structure, variance of gray scale distribution, orientation of local texture and directionality. We hypothesized that Morphological Component Analysis (MCA) based on sparse coding and the Sobel filter post-processing provides a possible texture decomposition for under-mm-range crack detection. In our experiments, decomposed images into their coarse and fine characteristics components were successfully obtained by using two dictionaries dual tree complex wavelet transform (DTCWT) and discrete wavelet transform (DWT) respectively and the coarse component was treated by the Sobel filter to exhibit the crack in a fine way. Our results demonstrated that the proposed MCA framework with the Sobel operator may contribute for an automation of crack detections even in open field severe conditions such as bridge decks.