AutoTracker: A Traffic Sign Change Detection System

The deployment of Internet of Things and cyber-physical technologies leads to increased collection of spatiotemporal data. In the transportation domain, this data is being used in an expanding range of applications, e.g., to enhance road safety. The capture of changes to traffic signs, e.g., due to wear, road construction, accidents, or theft, helps ensure well-functioning and safe road networks. However, this process has so far been largely manual, making it costly and meaning that changes are often captured with considerable delays. To capture changes more cost-effectively and more frequently, we introduce AutoTracker, an automated reporting system that exploits low-cost cameras in vehicles to capture changes. The system features a three-layer architecture—perception, analysis, and storage—that incorporates vehicle motion dynamics for improved performance. The analysis layer integrates physical modeling, R-tree-based retrieval, and calibration components to refine detection results. We also introduce new datasets and evaluation metrics to assess the performance of AutoTracker. Experiments show that AutoTracker is capable of state-of-the-art accuracy and efficiency in traffic sign change reporting.