An Automatic Live Load Survey Method based on Multi-Source Internet Data and Computer Vision

The measured live load forms the data basis for the reliability analyses. This study focuses on the amplitude measurement of the sustained load, which is an essential component of the live load. Traditional survey methods are characterized by manual and on-site operation, which can lead to a series of problems including the high cost, low efficiency and occupant resistance. Taking full advantages of the unlimited Internet resources and computer vision technology, a new survey method is proposed to realize an automatic and online investigation into the load amplitude. The amplitude statistics are derived from the survey data on the object weights, room areas and object quantities. Specifically, the object weights and room areas are directly acquired from the product information on e-commerce websites and the residence information on real estate websites, respectively. The object quantities are identified from the room photos on real estate websites. Therefore, an object detection model based on the YOLOv4 algorithm is developed. The load investigation into living rooms is used for illustrating the implementation process of the proposed method. The result of a previous survey covering 20040 m² suggests that 6 types of indoor objects contribute the majority of the load statistics and require to be considered in the detection model. The training, validation and test dataset include 5979, 1000 and 1000 room photos, respectively. The detection model has mean average precision (mAP) of 62% on the test dataset. For comparison, object quantities in 343 living rooms are obtained by both the manual counting and computer vision. The difference between the manual and automatic survey results is smaller than 20%, which verifies the feasibility and accuracy of the proposed method.

Keywords: Live load, Multi-source data, Computer vision, Automatic survey, Online method.

Download PDF