Distributed Acoustic Sensor System for Vehicle Occupancy Detection

A distributed acoustic sensors (DAS) system is investigated in this paper to offer vehicle occupancy detection for intelligent transportation systems (ITS). Accurate detection of the number of occupants in a vehicle is critical for ITS to mitigate safety risks, including monitoring overloading behaviour and ensuring the legal use of high occupancy vehicle (HOV) lanes. Several deep learning (DL) algorithms have been designed to obtain occupancy by analyzing DAS signals. An AlexNet and a slim visual geometry group (SVGG) algorithm are designed to analyze the data of a single DAS sensor. Furthermore, a sparse residual network (SR-Net) is proposed to process merged signals collected by multiple DAS sensors. We collected DAS data in road experiments and tested the proposed DL approaches. The test results indicate that 1) both the proposed Alexnet and SR-Net approaches can achieve detection accuracy of over 90%; 2) the proposed SR-Net model exhibits superior convergence, stability, and accuracy than the conventional residual network (ResNet); 3) the proposed DL solutions have practical complexity and model size.