This is repository of paper "Advancing construction site workforce safety monitoring through BIM and computer vision integration".

The project combines Building Information Modeling (BIM) and computer vision technology to facilitate both real-time and non-real-time monitoring of construction sites.

The system implementation contains 3 main modules:

vision_module/run.py contains computer vision related module for data collection:

• Object detection and classification - YOLOv8
• Object tracking - SORT
• Perspective projection
• Store data into database

In this step, we utilized Sqlite to store the data collected from computer vision module. The database named as data.db and contains two tables:

• temp_data - for real-time monitoring
• history_data - for non-real-time monitoring

Both of the tables are formatted as follows:

• person_id (int): object tracker id
• cam_id (int): camera id
• floor (int): camera location floor
• datetime (varchar): data collection time
• x_location (int): BIM x coordinate after projection
• y_location (int): BIM y coordinate after projection
• classification (int): worker safety catogory

This study utilizes Autodesk Revit for the BIM module. Revit includes a built-in visual programming interface, Dynamo, which we employed for preprocessing data from the database and subsequent visualization.

The images of the Dynamo code can be viewed at:

• BIM_module/DYNAMO_real_time_data_visualization.png for real-time data
• BIM_module/DYNAMO_historical_data_visualization.png for non-real-time data

The object detection dataset contains 3,455 images and randomly split into training, validation, and testing datasets with ratios of 83, 12, and 5%.

The labeling class includes:

• w - worker only
• wh - worker using helmet
• wv - worker using vest
• whv - worker using helmet and vest

Object detection accuracy

Projection accuracy

Mean error distance (MED) = 13.2 cm

@article{kulinan2024advancing,
  title={Advancing construction site workforce safety monitoring through BIM and computer vision integration},
  author={Kulinan, Almo Senja and Park, Minsoo and Aung, Pa Pa Win and Cha, Gichun and Park, Seunghee},
  journal={Automation in Construction},
  volume={158},
  pages={105227},
  year={2024},
  publisher={Elsevier}
}