Multi-Task Intelligent Monitoring of Construction Safety Based on Computer Vision

Traditional methods for ensuring construction safety are often manual, inefficient, and struggle to cover the comprehensive needs of dynamic construction environments. Although computer vision is being adopted, existing research largely focuses on single tasks, which limits its effectiveness. To address this, the study introduces a multi-task computer vision technology for improved surveillance. The methodology involves processing multi-source video data and adapting the YOLOv8 deep learning model through head component modifications. This enables the model to perform multiple functions concurrently, including detecting and segmenting construction-related objects, as well as estimating the poses of individuals and machinery. These functions are then integrated with a tracking algorithm to ensure continuous monitoring and facilitate the early detection of potentially hazardous situations. Furthermore, the paper highlights the creation of a new Integrated Excavator Pose (IEP) dataset, which helps overcome the limitations of isolated datasets and ensures the robust application of the proposed system in practical construction scenarios.