Development of a robust hazardous area tracking algorithm for pan-tilt-zoom (PTZ) cameras at construction sites under harsh conditions

Purpose Construction sites pose significant safety risks due to static hazardous areas and unauthorized worker intrusions. Conventional fixed-camera monitoring systems have limitations in tracking dynamic site conditions. This study aims to develop a robust hazardous area tracking algorithm using pan-tilt-zoom (PTZ) cameras, enabling real-time monitoring without additional sensors. Design/methodology/approach Based on homography, four approaches with detailed parameters are presented. To evaluate the performance of the four approaches, the experiment was conducted using both the virtual and actual construction sites of the Unity game engine. This system was tested by assuming various scenarios such as extreme camera movement, occlusion and bad weather. Findings Through the experiment, it was confirmed that the best performance was achieved by applying a deraining model and combining reference and previous-frame conversions based on the visual place recognition (VPR) results. Experimental results demonstrated high tracking accuracy, achieving an Intersection over Union (IoU) score, a standard metric for polygon segmentation accuracy in images, of 0.9555 in simulations and 0.8876 in real-world applications. Originality/value This study introduces a novel approach to hazardous area tracking by integrating deep learning-based homography estimation with PTZ camera systems, eliminating reliance on fixed viewpoints or external sensors. The proposed framework enhances proactive safety management by dynamically adapting to changing site conditions, improving detection reliability and reducing operational constraints.