Machine Learning Combined with Thresholding - A Blended Approach to Potholes Detection

Abstract

Potholes are a regular occurrence that can cause discomfort and harm everyday road users. In recent times many studies have been done on automated pothole detection as there is a need to assess the road condition in a more affordable and timely manner. This research aims to explore the different motion-based approaches used in pothole detection. Motion sensors such as accelerometers and gyroscopes are commonly utilised to acquire movement information, and these data can be used not only to detect the presence of potholes but also have been used to classify general road conditions. It has been found that the approaches can be divided into two categories: threshold-based and machine learning. For both approaches, statistical features are extracted from the motion data and used in determining the threshold values or as inputs to train the classifier models. Further opportunities for improvement in data labelling and the need to classify pothole severity levels using a standard metric are also discussed in the paper.