XClose

UCL Institute for Environmental Design and Engineering

Home
Menu

Seunghyeon Wang

Development of occupancy model using DANN and SIMS4 in office for privacy protection

HVAC (Heating, Ventilation, and Air-Conditioning) and lighting system are main energy end use in office. One way to save the energy consumption is to introduce automation saving system with occupancy model using camera. A difficulty of introducing camera can be to reveal occupant privacy during labeling process. Existing research tries to only make appearance of occupant identification more difficult. However, they are still required to compare original real image with modified image or re-identified through image resolution techniques such as 3D reconstruction and SRGAN (Super Resolution Generative Adversarial Network).

The main aim in this research is to propose occupancy modeling using DANN which does not require labeling of real occupant data for solving privacy problem caused by labeling process. For the implementation of this model, 1) Using SIMS4 is proposed as source data in DANN. The advantage of SIMS4 is to artificially make the corresponding similar environment to real office including occluded and crowded situations without privacy problem. 2) The architecture of DANN based on edge subtraction is suggested to reduce explicitly the gap of saturation between SIMS4 and real occupant. 3) Possibility of using the constructed SIMS4 as labeled source data to other office with different environment is validated for solving not only privacy problem but also labeling price.