Multisource Data Integration and Computer Vision Technology in Uncertainty Quantification of Live Loads

Abstract

Quantifying the uncertainty of live loads holds a significant reference value for reliability assessments at the component, structural, and even urban levels. Traditional survey methods are characterized by high survey costs, substantial manpower requirements, long implementation periods, and slow data updates. To address these shortcomings, this study proposes a new survey method that integrates heterogeneous data from various sources such as real estate and e-commerce websites, including photos of the surveyed region, object weights within the region, ownership change history, and so on. An object detection model is established using the You Only Look Once (YOLO) v4 algorithm. The model achieves mean average precision of 76% on the test data set and is applied to automatically identify object quantities from photos of the surveyed region. The feasibility and accuracy of the proposed method are verified through an illustrative survey example. Subsequently, this new method is applied to a large-scale survey in Shanghai, China, covering around 300,000 m2. Through the analysis of survey results, it was found that significant variations exist in the statistical outcomes between different districts within the same city. Specifically, the differences in statistical results for the load amplitude and change interval can reach 65% and 36%, respectively.