Traffic Intensity Detection Using General-Purpose Sensing

The conventional approaches for traffic intensity (TI) detection in smart cities use specialized sensors, such as loop detectors, cameras, and radar. These sensors come with high costs, limited reuse, and specialized maintenance. Thus, we propose the utilization of general-purpose sensing, which involves the use of cost-effective and easily deployable sensors, such as microphones, and air quality sensors that are cheaper, reusable, and easily manageable sensors. A general-purpose sensing infrastructure can be used for several applications including measuring TI. The main objective of this letter is to demonstrate how noise signatures can be leveraged to measure TI. Traditionally, image classification techniques are used for audio data analysis and vision-transformers (ViT) have shown remarkable results in this area. However, there is limited research that explores ViT models for traffic intensity detection. For TI detection, two approaches: vehicle count prediction and vehicle type detection (VTD), which use ViT models, are proposed. VTD approach performed better and it achieved an F1-score of 0.95 and a mean vehicle counting error of 0.032. In addition, the computational complexity of the approach was evaluated by implementing an edge solution. The proposed approaches can be effectively used even on resource-limited edge devices, with a notable increase in detection time.