Abstract
Occupancy information plays a key role in occupant-centric Heating, Ventilation, and Air Conditioning control and personalized thermal comfort. Recent advances in information technology and data-driven methods have increased interest in using computer vision for indoor occupancy detection. This paper presents a systematic review of computer vision-based occupancy detection approaches applied mainly to thermal environment control. The review analyses 48 papers published between 2020 and 2026, focusing on detected occupancy parameters, sensing modalities, deep learning models, dataset characteristics, evaluation metrics, and deployment limitations. The findings indicate that occupant count and activity are the most frequently addressed parameters, while clothing classification and skin temperature estimation are mainly associated with personalized comfort assessment. YOLO models dominate recent implementations and RGB images are the most widely used sensing modality despite growing interest in thermal and multimodal approaches. The review also highlights limitations, including inconsistent benchmarking, heterogeneous evaluation metrics, and limited attention to deployment constraints such as privacy, scalability, and integration with building management systems. Overall, computer vision shows strong potential for occupant-centric HVAC control, but broader adoption requires standardized evaluation, control-oriented validation, and deployment-aware reporting.