Advancing Vehicle Classification: A Novel Framework for Type, Model, and Fuel Identification Using Nonvisual Sensor Systems for Seamless Data Sharing

Vehicle classification (VC) down to type and model is exclusively visual-/image-based and struggles with restricted data availability and latency in data-sharing toward an increasing market of public and private mobility service providers. This article presents a framework architecture for VC down to type and model using a microelectromechanical systems (MEMS) intrusive magnetic sensor system combined with advanced machine learning (ML). The framework follows the principles of “privacy-by-design” (PbD) and adds fuel identification as a new attribute. The framework was validated in an experiment on nine predefined passenger car models, confirming the capability of nonvisual sensors to classify vehicle type, model, and fuel. This innovative approach has the potential to significantly impact how transport and infrastructure agencies collect, process, and share intelligent traffic monitoring systems (ITMS) data in the future, providing more accurate and detailed information for improving roadway utilization efficiency, predicting transportation needs, enhancing transportation safety, and reducing environmental impact. The work proposes a system framework, experimental design, and evaluation against automatic number plate recognition. The evaluation involved collaboration among various partners, including academia, public road administration, and private companies, with diverse interests that need to be preserved due to significant development progress.