The Washington State Department of Transportation (WSDOT) faces challenges in accurately monitoring traffic across its extensive roadway network due to the limited deployment of costly Permanent Traffic Recorders (PTRs) and traditional traffic sensors. These sensors, including inductance loops, video image processors, and microwave sensors, cannot meet the Federal Highway Administration's (FHWA) 13-category vehicle classification requirements with sufficient accuracy. With less than 200 PTRs deployed statewide, large portions of rural highways remain unmonitored, and short-term manual data collection methods pose safety risks for field staff at sites with challenging geometry and high traffic volumes.
To address these limitations, this research introduces a cost-effective, machine learning (ML)-driven traffic data collection system leveraging Mobile Unit for Sensing Traffic (MUST) video cameras. By employing deep learning techniques for object detection, classification, localization, tracking, and counting, the system can achieve high-accuracy 13-category vehicle classification. Additionally, the integration of edge-based data analysis ensures real-time processing without relying on communication infrastructure, making it ideal for remote and safety-critical locations.
Figure 1. AI-powered vehicle classification using MUST video cameras for WSDOT, accurately detecting and categorizing vehicles into 13 FHWA classes with confidence scores.