Vehicle Detection and Classification by Voiceprint Recognition Based on Single Acoustic Sensor Under Bridge Expansion Joint
Yue Pan, Yiqing Dong, Dalei Wang, Jin Di, Airong Chen
Abstract
Vehicle detection and classification (VDC) is a crucial aspect of bridge engineering, as vehicles exert significant dynamic loads on bridges. Various contact and noncontact methods have been proposed for VDC, but finding a durable and cost-effective sensing approach remains challenging for practical bridge applications. In this study, we propose a voiceprint recognition (VPR) method for VDC using a single microphone, offering flexibility and affordability. The distinctive acoustic signatures generated by vehicles impacting bridge expansion joints (BEJs) are captured and utilized for VDC. We optimize a threshold-based algorithm for vehicle detection and a deep learning-based VPR model for vehicle classification. In addition, cascading VPR models enable fine-grained vehicle classification, including lanes and axle types. We validate the proposed method on an actual bridge. The short-term energy (STE) thresholding algorithm achieves a detection accuracy and recall of 90.0% and 91.6%, respectively. The ConFormer model achieves an area under the curve (AuC) of 0.925 for vehicle classification. These results highlight the method as an easy-to-implement and cost-effective solution for efficient VDC surveys. Future work can focus on expanding datasets and incorporating multiple microphones to further enhance the system’s capabilities.