Distributed sensing vehicle flow detection method based on deep learning

This article proposes a traffic detection method based on deep learning and distributed acoustic sensing (DAS). Mainly based on the improved YOLOv8 algorithm for trajectory recognition, traffic flow counting is performed based on the recognition results, and DeepSORT algorithm is used for trajectory tracking to facilitate comparison of traffic flow. This article uses data collected by DAS systems in real roads for data preprocessing and training. The collected data is converted from text data to waterfall plots for processing. Data preprocessing involves normalizing the collected data and using an improved wavelet threshold denoising algorithm. In addition, a comparison was made between the improved model and traditional methods, and the results showed that using the improved YOLOv8 algorithm can obtain more accurate recognition results and apply them to subsequent counting and tracking work, which can more accurately calculate traffic flow information. This has a good effect on traffic situation estimation and prediction in the field of transportation.