Abstract
Ubiquitously deployed Internet of Things (IoT)-based automatic vehicle classification systems will catalyze data-driven traffic flow optimization in future smart cities and will transform the road infrastructure itself into a dynamically sensing Cyber-physical System (CPS). Although a wide range of different traffic sensing systems has been proposed, the existing solutions are not yet able to simultaneously satisfy the multitude of requirements, e.g., accuracy, robustness, cost-efficiency, and privacy preservation. In this paper, we present a novel approach, which exploits radio fingerprints—multidimensional attenuation patterns of wireless signals—for accurate and robust vehicle detection and classification. The proposed system can be deployed in a highly cost-efficient manner as it relies on off-the-shelf embedded devices which are installed into existing delineator posts. In a comprehensive field evaluation campaign, the performance of the radio fingerprinting-based approach is analyzed within an experimental live deployment on a German highway, where it is able to achieve a binary classification success ratio of more than 99% and an overall accuracy of 93.83% for a classification task with seven different classes.
Introduction
Optimizing traffic in future smart cities requires granular data on vehicle types to enable advanced lane management, tolling, and flow optimization. However, current sensing technologies face significant trade-offs. Intrusive sensors like inductive loops require expensive roadwork for installation and maintenance. Vision-based systems raise privacy concerns and struggle in adverse weather conditions like heavy rain or fog. Meanwhile, high-accuracy solutions like LiDAR and Radar are often too costly for ubiquitous deployment.
This research proposes a "Goldilocks" solution: utilizing the existing radio environment as a sensor. By analyzing how vehicles disrupt wireless signals, the system provides a low-cost, weather-robust, and privacy-preserving alternative that transforms standard road infrastructure into a sensing network.
Methodology and System Design
The core of the proposed system is Radio Fingerprinting. When a vehicle passes between a radio transmitter and receiver, its physical geometry—size, shape, and material—obstructs the signal, creating a unique multidimensional attenuation pattern.
Hardware and Deployment
The system utilizes off-the-shelf, low-power IEEE 802.15.4 radio modules operating at 2.4 GHz. These are installed inside standard roadside delineator posts, making the system nearly invisible and easy to deploy. For a single-lane setup, three transmitters are placed on one side of the road and three receivers on the other, creating a mesh of nine distinct radio links.
Signal Processing and Classification
The system samples the Received Signal Strength Indicator (RSSI) every 8ms. A moving average filter cleans the raw data, and a threshold-based state machine detects vehicle entry and exit. To classify these fingerprints, the authors evaluated several machine learning models, including Random Forests and Support Vector Machines (SVM). A linear SVM was ultimately selected for its optimal balance of high accuracy and low computational overhead, making it suitable for real-time edge computing on microcontrollers.
The Concept of Radio Fingerprinting
The unique "shadow" cast by a vehicle in the radio field serves as its signature. As shown in the RSSI time series below, different vehicle classes produce markedly different attenuation profiles.
A passenger car typically causes a drop of ~25 dB with a relatively short duration, whereas a larger truck causes a deeper drop (~35 dB) and a significantly longer duration with higher signal variance. By leveraging nine overlapping links, the system gains redundancy and robustness against noise.
Experimental Evaluation and Results
The system was evaluated using a real-world dataset collected from an experimental deployment on a ramp of the German highway A9. The dataset comprised fingerprints from 2,605 vehicles, with ground truth provided by synchronized camera images.
- Detection Performance: The system achieved a 100% detection rate and correctly identified driving direction in all cases.
- Classification Accuracy: Binary classification (car vs. truck) reached over 99% accuracy. Fine-grained classification across seven body styles (e.g., vans, buses, semitrucks) achieved 93.83% accuracy.
- Computational Efficiency: The linear SVM model is extremely lightweight, requiring only 0.22 ms for classification on an ESP32 microcontroller. This enables real-time processing at the roadside using battery-operated IoT devices.
Conclusion and Future Work
Radio fingerprinting offers a promising path toward ubiquitous traffic sensing. It is cost-effective, privacy-preserving, and highly accurate. While the current study focused on single-lane scenarios, future work will address multi-lane environments and explore the use of millimeter-wave (mmWave) technology to further improve resolution and mitigate interference. Addressing class imbalance for rarer vehicle types, such as buses, also remains a priority for future model refinement.
Summary updated