Enhancing traffic monitoring with noise-robust distributed acoustic sensing and deep learning

Traffic monitoring provides crucial data for intelligent transportation systems (ITS) but traditional sensors are expensive to deploy and maintain at scale. This study explores distributed acoustic sensing (DAS) using existing fiber-optic infrastructure as a cost-effective solution for traffic monitoring. While DAS offers advantages, vehicle detection signals are susceptible to noise. To address this, we propose a novel approach combining DAS with deep learning object detection using YOLOv8. Pre-processed and labeled DAS data collected over two weeks on a highway during a COVID-19 lockdown were used to train the YOLOv8 network, achieving 92 % classification accuracy. Applying the trained model revealed detailed hourly traffic patterns and vehicle compositions, demonstrating the potential of DAS for robust and cost-effective ITS. These findings highlight the effectiveness of combining DAS and deep learning for noise mitigation in traffic monitoring and provide valuable insights into traffic dynamics during the pandemic.