An Edge Ai and Computer Vision Method For Understanding Human Behavior In Architecture

This study evaluates the performance of edge AI computer vision models for detecting human occupants in indoor architectural spaces. Six pre-trained models were implemented on low-cost single-board computers and tested using offline video footage. Performance was assessed based on computational efficiency and detection accuracy. Results show acceptable performance levels, particularly with YOLO and EfficientDet models. While some challenges remain, such as unusual perspectives and occlusions, the findings confirm the viability of edge AI computer vision for anonymous, real-time occupant detection in buildings. This enables large-scale, continuous analysis of human behavior in architectural spaces without compromising privacy. The study concludes that edge AI computer vision offers significant potential for data-driven building operation optimization and design feedback, opening new possibilities for creating spaces that better respond to actual human needs and behaviors.