CLASSIFICATION OF ROAD SURFACE QUALITY USING ANDROID SMARTPHONE DEVICES

Abstract The quality of roads in a country contributes greatly to its economic development. In Ghana and elsewhere in Africa, essential goods such as agricultural produce are transported mostly by road. Good roads not only promote economic activity, they contribute positively to the quality of life of their users. In many areas, however, the road infrastructure is not of uniform quality, comprising tarred roads, dirt roads, smooth roads, bumpy roads, and roads that are barely motorable. Roads with poor surface conditions damage vehicles, slow down traffic, lead to accidents, and are uncomfortable to drive on. A possible solution to the problem is system that automatically detects and reports the surface conditions of roads. This study explores the use of an Android-based mobile application to detect and report the surface quality of roads. We start by collecting hand-labeled training data from vehicles traversing carefully selected roads of differing quality. In addition, we use GPS sensors to reliably match road quality information to specific locations. A logistic regression machine learning algorithm is used to train a road surface classifier based on the accelerometer readings collected. This study aims to detect the surface condition of roads and present that data in a manner that can be easily be embedded into maps online. We find that we are able to distinguish between good and bad roads with a true positive rate of 92%. We are able to distinguish between good and fair roads with a true positive rate of 83%. The study is however unable to reliably distinguish between fair and bad roads. To the best of our knowledge this study is the first to attempt the automatic classification of entire sections of road as opposed to the detection of individual road anomalies such as potholes.