Road Crack Detection Using Gaussian Mixture Model for Diverse Illumination Images

The studies on road crack detection using machine learning techniques have been conducted and derived significant accuracies. Most of the studies do not consider diverse illumination during daytime or nighttime, and only focus on the detection of category or presence of road crack. In this paper, we propose IlumiCrack, a novel road crack detection framework collaborating Gaussian mixture model (GMM) and object detection CNN models to address these issues. The contributions of this paper are: 1) For the first time, a large- scale road crack dataset with a variety of illumination such as the pictures at nighttime are prepared using a dashcam. 2) On the basis of GMM, experimental evaluations on 2 to 4 levels of brightness classification is conducted for optimal classification. Experimental results show that IlumiCrack outperforms the state-of-the-art R-CNN object detection framework.