Worker Identity–Aware Tracking across Multiple Cameras Using Multimodal Data Fusion Techniques

Abstract

Accurate identification and tracking of construction workers is critical for real-time safety and productivity management on complex and dynamic construction sites. However, traditional sensor-based wearable tools for worker identifying face issues such as physical interference of workers, the need for recharging, the need for maintenance, and high cost. Computer vision technique (CVT)-based methods are limited to merely reidentifying the same workers in a single or multiple cameras, rather than identifying who the workers are on the construction site. This study focuses on reidentifying who the workers are, namely worker identity–aware tracking, by using entrance access images as personal identity anchors and multimodal data fusion techniques. Personal images from the entrance access system are exploited as identity anchors to be matched with personal images from multiple cameras on site, serving as personalized prototypes that are updated dynamically to maintain robustness under varying imaging conditions and temporal appearance changes. In addition, personal protective equipment (PPE) information contained in the entrance access records is integrated to enhance discriminability among workers with similar appearances. The proposed identity-aware tracking method was implemented, tested, and applied to a case construction site with multiple surveillance cameras and real operational scenarios. The proposed method achieved more stable and reliable identity association than approaches that match incoming images against a global gallery. This study provides domain knowledge of a nonintrusive, cost-effective and intelligent methodology for worker identity–aware tracking on construction sites. It also provides a digital platform to help achieve remote, timely, precise, and effective field management of workers, thus promoting construction productivity and safety.