Toward Community Sensing of Road Anomalies Using Monocular Vision

Abstract

Advanced vehicle safety is an emerging issue appealed from the rapidly explosive population of car owners. Posing a remarkable safety threat, road anomalies not only damage vehicles but may also cause serious danger, especially at night or under bad visibility conditions. However, maintaining the quality of roadways has been a big challenge for municipalities around the world. Recently, the rapid development and reduced cost of digital cameras have made it economically feasible to deploy driving video recorders (DVRs) on vehicles. Thus, in this paper, we employ the widespread DVRs as distributed sensors with high mobility to conduct pervasive sensing of road anomalies. First, vehicle shakes are detected to infer the candidates of road anomalies. Then, we segment pavement regions, extract saliencies on the road surface, and classify whether a detected vehicle shake is caused by a road anomaly or an artificial speed bump. Experiments are conducted on a test data set collected by front-mounted DVRs, and the results verify that the proposed system can effectively detect road anomalies in real time, showing its good feasibility in real-world environments.