| Abstract |
The Internet of Things (IoT) infrastructure, systems, and applications demonstrate potential in serving smart city development. Crowdsensing approaches for road surface conditions monitoring can benefit smart city road information services. Deteriorated roads induce vehicle damage, traffic congestion, and driver discomfort which influence traffic management. In this paper, we propose a framework for monitoring road surface anomalies. We analyze the common road surface types and irregularities as well as their impact on vehicle motion. In addition to the traditional use of sensors available in smart devices, we utilize the vehicle motion sensors (accelerometers and gyroscopes) presently available in most land vehicles. Various land vehicles were used in this paper, spanning different sizes, and year model for extensive road experiments. These trajectories were used to collect and build multiple labeled data sets that were used in the system structure. In order to enhance the performance of the sensor measurements, wavelet packet de-noising is used in this paper to enable efficient classification of road surface anomalies. We adopt statistical, time domain, and frequency domain features to distinguish different road anomalies. The descriptive data sets collected in this paper are used to build, train, and test a system classifier through machine learning techniques to detect and categorize multiple road anomalies with different severity levels. Furthermore, we analyze and assess the capabilities of the smart devices and the other vehicle motion sensors to accurately geo-reference the road surface anomalies. Several road test experiments examine the benefits and assess the performance of the proposed architecture. |
