PROJECT TITLE :

BARNet Boundary Aware Refinement Network for Crack Detection

ABSTRACT:

One of the most noticeable issues that can frequently manifest itself in highways and main roads is the formation of road cracks. The manual evaluation of road cracks is laborious, takes a lot of time, and can be inaccurate. In addition, it has a number of problems with the way it is implemented. On the other hand, the computer vision-based solution is extremely difficult to implement because of the complex ambient conditions, which include illumination, shadow, dust, and the shape of the crack. The majority of the cracks appear as irregular edge patterns, which are the most significant characteristics for the purpose of detection. The most recent developments in deep learning utilize a convolutional neural network as the foundational model to detect and localize cracks using only a single RGB image. However, this method produces edges that are fuzzier and more pronounced because the boundary it uses for crack localization is inaccurate. In order to solve this issue, the research presents a novel and reliable method for detecting road cracks that is based on deep learning and also takes into account the image's initial edge as an additional feature. This method is intended to help overcome the challenge. The primary contribution of this work is the modification of the initial image gradient using the coarse crack detection result to produce more precise crack boundaries. This modification was accomplished by refining the original gradient. Extensive experimental findings have demonstrated that the proposed method outperforms the previous methods that were considered to be state-of-the-art in terms of accuracy of detection.