Construction sites pose a high risk of struck-by accidents due to limited situational awareness. This study investigates the feasibility of using a neckband-style wearable 360° camera for proximity warning in construction environments. The proposed framework consists of a data preprocessing module—handling motion blur detection and distortion correction—and a hazard detection module, which integrates object detection with depth estimation. By identifying hazardous equipment and estimating its distance, the system provides personalized warnings to workers at risk. Field tests conducted at actual construction sites demonstrated an F1-score of 0.81 in hazard detection. Unlike fixed cameras or sensor-based systems, this approach enables omnidirectional, individualized hazard monitoring without additional site infrastructure. The wearable camera also minimizes interference with tasks, supporting stable image capture. The findings highlight the potential of combining wearable computer vision and deep learning to improve safety in complex and dynamic construction settings.
Yoo et al. (Thu,) studied this question.