LAUSR

Road anomaly detection with crowdsourced sensor data has become an increasingly important field of research over the last few years. Traditional ways for road anomaly detection are either threshold-based detection techniques or feature-based detection techniques. However, road anomaly patterns from crowdsourcing data are often shifted in time and exhibit local distortions/noise, thus existing methods rely on the original sensor data greatly limit the accuracy of road anomaly detection. In this paper, we present a road anomaly detection model by learning scale-invariant features from the differences between small local segments of road anomaly samples. Specifically, the proposed model consists of two phases: 1) Road anomaly segmentation. The phase is designed to roughly extract road anomaly subsequence using piecewise aggregate approximation representation of sensor series data, and 2) Road anomaly detection. In this phase, we observe the differences among road anomaly classes are attributed to small local segments, then we learn scale-invariant features from these small local segments for road anomaly detection. To demonstrate the utility of our proposed model, we have performed a comprehensive experimental evaluation on two real-world datasets and one large-scale simulation dataset. The experimental results show our proposed model outperforms all baselines significantly in terms of road anomaly detection.