Identification of the spatiotemporal location of vehicle loads on highway bridges based on multi-view information fusion

Abstract

To solve the problem that existing methods have difficulty in accurately obtaining the spatiotemporal distribution of vehicle loads on bridges in complicated traffic scenes, a spatiotemporal location identification method for vehicle loads based on multi-view information fusion is proposed. First, the vadYOLO-StrongSORT model is developed to detect and track vehicles simultaneously in a single view. Furthermore, based on image calibration and cross-view vehicle matching, an adaptive weighted least squares method is used for multi-view information fusion to correct the vehicle trajectory. Finally, the spatiotemporal distribution of axle loads is reconstructed by combining vehicle trajectories with axle configurations. The performance of the proposed method under typical traffic conditions is evaluated using model tests. The results show that the multi-view information fusion method significantly improves tracking stability, localization accuracy, and anti-occlusion performance compared with the single view-based vehicle location identification method. In the lane-changing scenes, the highest average localization error of the proposed method is less than 2.0 cm, which is significantly better than the 17.0 cm of the single-view method. In multivehicle occlusion scenes, the proposed method achieves a vehicle capture rate of up to 100%, compared with a maximum of only 72.5% for the single-view method. Meanwhile, vadYOLO-StrongSORT achieves the highest identification accuracy in the experiment compared with other detection and tracking models.

Keywords

bridge engineering / vehicle loads / spatiotemporal location / multi-view information fusion / vehicle axle identification / bridge weigh-in-motion / bridge health monitoring