Estimating Worker-Centric 3D Spatial Crowdedness for Construction Safety Management Using a Single 2D Camera

Abstract

As a major risk factor that leads to struck-by accidents, crowdedness indicates the number of workers within the range of a targeted worker (this range varies according to the construction site). High crowdedness can result in dangerous working conditions, negative workers’ behaviors, lack of concern for safety climate, and productivity loss due to saturated and insufficient working areas where workers can perform. An automatic computer vision-based technique could be a novel solution for crowdedness monitoring for proactive safety management. Non-intrusiveness and applicability in a complex outdoor environment are critical considerations for device selection on construction sites. Accordingly, a red, green, and blue (RGB) camera is selected to detect worker-centric crowdedness. This device is less intrusive for workers than wearable sensors and is also widely applied in outdoor construction sites considering complex working areas and various light conditions. Previous RGB camera-based methods for crowdedness detection simplify the proximity estimation process by assuming that the construction site is a two-dimensional (2D) planar surface. These methods use image pixels for proximity calculation. Such simplification can cause a distortion in three-dimensional (3D) spatial proximity due to 2D projection of 3D entities. Moreover, previous methods suffer from lack of reproducibility due to the view variance of a 2D camera. To address these problems, a 3D spatial crowdedness estimation method is developed by generating a 3D space for proximity and crowdedness calculation from 2D video frames. This method has been validated in laboratory and field tests. Results indicate that the proposed method enables the estimation of 3D spatial proximity between two workers within an error of 0.45 m in a real-time and view-invariant manner from a 2D video. The proposed method is expected to enable managers to accurately monitor crowdedness among workers for proactive construction safety management.