In this paper, a method for constructing and verifying datasets used in deep learning technology is proposed to prevent safety accidents in the construction environment. The virtual construction simulator was developed to implement various constructio...
In this paper, a method for constructing and verifying datasets used in deep learning technology is proposed to prevent safety accidents in the construction environment. The virtual construction simulator was developed to implement various construction environments, and custom data in the virtual environment was set to diversify the location and parameters of the sensors. A database for real-time deep learning models was built and detected at the pixel level to detect uneven roads and objects in the construction site. When deep learning-based detection is performed, it may interfere with the accurate judgment due to occlusion areas hidden by other objects. Therefore, the database for developing the occlusion area detection algorithm in virtual environments was constructed and verified.