Improving metabolic rate estimation in buildings through diffusion-based occluded pose reconstruction

School of Architecture, University of Southern California, Los Angeles, CA 90089, USA

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Human-integrated building systems increasingly rely on accurate recognition of occupant activities to support thermal comfort assessment and energy-efficient HVAC control. Vision-based activity recognition using indoor images has shown significant potential for real-time estimation of occupant activity levels; however, its performance is often degraded by occlusion caused by furniture, surrounding objects, or occupant postures in indoor environments. In particular, such occlusion leads to missing body information, resulting in unstable pose estimation and reduced accuracy in metabolic rate estimation. To address this limitation, this study proposes an occluded pose reconstruction framework that enhances vision-based activity recognition by reconstructing missing body regions in occluded indoor images. The proposed framework explicitly masks occluded regions and selectively regenerates them using diffusion-based image reconstruction, enabling more stable pose estimation and improving the accuracy of metabolic rate estimation. The performance of the proposed approach was evaluated using indoor activity images under two levels of occlusion through an analysis of reconstruction consistency and a comparison of metabolic rate estimation accuracy. The results demonstrate that the proposed reconstruction-based approach improves activity recognition and metabolic rate estimation by more than 10% under occlusion, particularly for sitting activities in high-occlusion conditions and standing activities in low-occlusion scenarios. The findings further reveal that the effectiveness of image reconstruction depends on both the level of occlusion and the type of activity, highlighting practical limitations under extreme occlusion conditions. These findings suggest the potential to support stable thermal environment control across a broader range of occlusion conditions, especially when image reconstruction is complemented by additional occlusion-aware mechanisms.