Pothole Detection with Image Segmentation for Advanced Driver Assisted Systems

It is common to see water trapped on roads after rain. Potholes are the main reason behind trapped rainwater on roads. In general, it is advisable to avoid these water pools when driving. This research presents a vision-based Advanced Driver Assistance System that detects the water pools on asphalt roads. The presented system will assist drivers to take evasive action. Since most of the automobiles come with a camera system, we propose image-based simple machine learning architecture to detect water pools. The paper analyses the contribution of edge features, texture features and Gabor filtered features for the detection of the water surfaces. Initially, these features are ranked according to their importance. Next three Random Forest Classifier models are created and trained using a random number of features from the feature set according to their ranks. Training accuracy of all 3 models varies from 0.94-0.99 indicating a promising outlook for the given application.