Prediction of human-induced structural vibration using multi-view markerless 3D gait reconstruction and an enhanced bipedal human-structure interaction model

In the context of advancing material engineering and construction technology, structures are evolving to be lightweight, giving rise to a heightened focus on human-induced vibration serviceability. Despite the availability of various Human-Structure Interaction (HSI) models, integrating outdoor tests with these models remains challenging due to the lack of a comprehensive testing framework. Existing methods heavily rely on invasive wearable sensors, lacking non-invasive alternatives. To bridge this gap, this paper proposed an outdoor testing framework for evaluating human-induced structural vibrations. Using a 2D body keypoints detection network, human gaits were captured from multiple viewpoints, representing it with the Skinned Multi-Person Linear Model (SMPL) model through triangulation and optimization. Gait and walking force data from 30 participants were analyzed using a Long Short-Term Memory (LSTM) network to classify landing states, which indicate whether both feet are in contact with the structure. Extending a bipedal HSI model from 1D to a 2D structure, walking tests were conducted on a 19.8 m × 2.35 m outdoor footbridge to update dynamic properties. Results showed over 90 % accuracy in predicting human landing states and within 10 % relative Root Mean Square Error (RMSE) in predicting pedestrian vertical walking force. Comparing models with and without HSI, disparities of 20 % to 60 % in frequency changes and 50 % to 180 % in damping ratio values were observed. The proposed non-invasive method predicted vertical structural vibration response with <10 % error, outperforming cases that used walking loads from force-measuring insoles without accounting for time-varying dynamics. These findings affirmed the feasibility and accuracy of our multi-view, non-invasive human gait acquisition method coupled with the improved bipedal HSI model in human-induced vibration prediction.