Smartphone-Based Pothole Detection Utilizing Artificial Neural Networks

Abstract

Roadway pavement maintenance to the preferred level of serviceability comprises one of the most challenging problems faced by civil and transportation engineers, with regard to transport infrastructure management. This paper presents a study on the detection of roadway pavement anomalies by use of smartphone sensors and on-board diagnostic (OBD-II) devices, which can lead to low-cost roadway infrastructure assessment. The proposed approach, which, in addition to smartphone sensors, also utilizes artificial neural network (ANN) techniques in the analysis, captures a vehicle’s interaction with a roadway pavement while the vehicle is moving, and utilizes the observed interaction patterns for the detection of potholes in the pavement. The method utilizes four metrics in the analysis and shows a detection accuracy of about 90%. Preliminary results on the inclusion of additional roadway defects in the analysis and on the ability of the method to distinguish between potholes and other pavement defects (e.g., patches, local upheavals, rutting, and corrugation) have been positive. The study’s results confirm the value of smartphone sensors in the low-cost (and eventually crowdsourced) detection of potholes.