Online anomalous vehicle detection at the edge using multidimensional SSA

We are witnessing the giant leap of Smart Cities and the prosperity of Internet of Things (loTs), but the anomalous vehicle surveillance issues in Intelligent Transportation Systems (ITS) are still challenging, especially when more computing tasks are migrated from Cloud center to the network edge. Different research studies attempted to tackle this problem, however, existing approaches are either requiring large training data sets or presenting detection unreliability without prior knowledge. In this paper, we propose to identify anomalous vehicles on roads in real-time using multidimensional Singular Spectrum Analysis (mSSA). Inspired by the excellent performance of the SSA algorithms in change point detection in time series, we adopted it to catch the differences in the dimension of characteristics of vehicles on roads. The multiple factors of vehicles' behavior are mapped into multiple channels in the mSSA framework. Instead of pre-training or defining normal motion patterns of vehicles, the anomaly detection is formatted as an outlier identifying problem. Using two vehicle trajectory data sets, the feasibility and effectiveness of our approach are verified. Comparing to other proposed methods like clustering, the experimental results show that our approach is more reliable and robust.