Traffic flow and vehicle speed monitoring with the object detection method from the roadside distributed acoustic sensing array

Abstract

Distributed acoustic sensing (DAS) is an emerging technology that transforms a typical glass telecommunications cable into a network of seismic sensors. DAS may, therefore, concurrently record the vibrations of passing vehicles over tens of kilometers and shows potential to monitor traffic at a low cost with minimal maintenance. With big-data DAS recording, automatically recognizing and tracking vehicles on the road in real time still presents numerous obstacles. Therefore, we present a deep learning technique based on the unified real-time object detection algorithm to estimate traffic flow and vehicle speed in DAS data and evaluate them along a 500-m fiber length in Beijing’s suburbs. We reconstructed the DAS recordings into 1-min temporal–spatial images over the fiber section and manually labeled about 10,000 images as vehicle passing or background noise. The precision to identify the passing cars can reach 95.9% after training. Based on the same DAS data, we compared the performance of our method to that of a beamforming technique, and the findings indicate that our method is significantly faster than the beamforming technique with equal performance. In addition, we examined the temporal traffic trend of the road segment and the classification of vehicles by weight.