A deep learning-based computer vision approach for crowd segmentation: a hospitality case study

Despite the proliferation of video analytics applications supported by deep learning (DL) methods in various service sectors, the potential of such technologies for hospitality decision support remains at an emergent stage. While issues have been persistent for researchers, this study identifies a real-world business problem in hospitality: the lack of insight into crowds (potential customers) needed to provide customized services and design promotional offers based on family demographics. To bridge this gap, through the lens of design science, this study introduces a DL-based video analytics solution (artifact) to detect and track potential customers, according to their family size, specifically identifying individuals, couples, and groups. Using 25 hours of CCTV footage, a DL model combining YOLOv11 and BYTEtrack was trained and developed (i.e. demonstrated as a software prototype) to detect and track individuals, couples, and groups. Initial evaluation results show strong performance, indicating that the model is suitable for further refinement and potential commercial application. This paper addresses a practical gap in hospitality operations by showing how a DL-based video analytics can generate insightful visual cues to support informed managerial decision-making.