Identification of Influence Lines for Highway Bridges Using Bayesian Parametric Estimation Based on Computer Vision Measurements
Yun Zhou, Jin-Nan Hu, Guanwang Hao, Zhengrong Zhu, Jian Zhang
Abstract
Conventional methods to identify influence lines, which are essential in design and evaluation of bridges, use contact sensors involving high upfront and operational costs. This paper presents an approach to identifying influence lines based on computer vision measurements. The approach integrates vision-based identification of vehicle types, estimation of vehicle loads, bridge displacement measurement, and Bayesian parametric estimation. A you only look once version 4 (YOLOv4)—a real-time object detector—with a convolutional block attention module is trained to identify vehicle types and estimate vehicle loads. Bridge displacement measurements provide dynamic deflections, which are then used to analyze the influence line through Bayesian parametric estimation. The performance of this approach was evaluated through laboratory and field experiments with different types of vehicles and driving speeds. The results show that the errors were up to 4.88% for laboratory experiments and up to 11.48% for field experiments. This research provides findings that will help with the practices of condition monitoring and assessment of highway bridges.