Using commercial floating car data to remotely infer the presence of potholes along rural road segments

Abstract

Potholes contribute to crashes, cause extensive damage to vehicles, and lead to further deterioration of road infrastructure. Remote detection of potholes is of great interest to travelers wishing to avoid pothole riddled routes, and to roads authorities for timeous protection of infrastructure. This study developed a method that can automatically and remotely infer that potholes exist along road segments using readily available traffic data. A simple Pothole Occurrence Probability (POP) Model is proposed that uses only commercial floating car data (FCD) as input. Commercial FCD are anonymized, widespread, and passively collected by GPS enabled devices, making FCD particularly appropriate for input to remote traffic monitoring. The application of FCD to infer pothole presence is unique and has not been previously investigated. Pothole presence is shown in this paper to significantly impact harmonic mean speeds reported by FCD along rural roads in South Africa. The relationship between pothole severity, evaluated from test-vehicle GPS data and dashboard-camera footage, and FCD-reported speed profiles, were empirically investigated along 69 km of training routes to develop the POP Model. The model was evaluated along six testing routes, with a total length of 189 km. 85 % of the testing routes were correctly categorized as either having or not having potholes, while 96 % of potholed road segments were correctly identified. The POP Model has wide application potential, firstly as input to navigation applications for travelers, and secondly through incorporation into pavement management systems to continuously monitor vast rural road networks.