Team of Researchers Awarded $1M Department of Energy Contract

in ECE - Data Science and Intelligent Systems, ECE - Mobile/Cloud Computing and Cybersecurity, ECE Spotlight Faculty, ECE Spotlight-Research, ME Spotlight Faculty, NEWS, SE Spotlight Faculty, SE Spotlight-Research

Project aimed at building a sensor system that will reduce energy costs in commercial buildings

A team of College of Engineering researchers has won a $1 million contract from the Department of Energy’s Advanced Research Projects Agency-Energy (ARPA-E) to develop COSSY (Computational Occupancy Sensing SYstem), a system of sensors that can estimate the number of people in a room and adjust air flow in heating, ventilation and air conditioning (HVAC) appropriately, with the goal of saving energy.

Professor Janusz Konrad (ECE) will lead the three-year project, one of 15 funded by ARPA-E and aimed at reducing HVAC energy usage, which accounts for 37 percent of energy consumed in commercial buildings in the United States.

Most commercial buildings operate air flow on a timer according to fixed schedule, typically a minimum air flow at night and another air flow matched to room’s capacity during the workday. With a system of sensors that can provide an accurate count of the number of people in a room, the workday air flow can be reduced while still maintaining air quality. By reducing the air flow across many rooms in a building, the team can meet the ARPA-E’s goal of 30 percent average energy savings.

The team is well equipped to attack this challenge: Konrad specializes in signal processing and computer vision; Professor Prakash Ishwar (ECE, SE) in machine learning; Professor Thomas Little (ECE, SE) in networking; and Associate Professor Michael Gevelber (ME, MSE, SE) in control of commercial HVAC systems and energy. Additionally, three industrial partners have already agreed to collaborate with the BU team on the project.

The team will develop a scalable system that uses both high-resolution panoramic cameras and low-resolution thermal door sensors. The cameras and door sensors will work together to estimate how many people are in the room at any given time. When visual privacy is of concern, just the thermal sensor will be installed so the system can still sense the number of people without using a camera.

A series of time-lapse thermal door sensor data shows a person entering a room without revealing their identity. The thermal sensors can still gather occupancy data and also protect privacy when needed.
A series of time-lapse thermal door sensor data shows a person entering a room without revealing their identity. The thermal sensors can still gather occupancy data and also protect privacy when needed.

The team will build COSSY to meet ARPA-E’s requirements for accuracy, security and cost. The primary focus will be on the development of algorithms to accurately estimate the number of occupants in a room from the visual and thermal data. Developing such algorithms becomes complicated when multiple sensors are needed for large or oddly shaped rooms. The sensors must communicate with each other to make sure people are not counted multiple times. If COSSY underestimates the occupancy, and the air flow is too low, air quality will suffer and people may feel discomfort. But if the estimate is too high, energy savings will not be optimized. By using state-of-the-art computer vision and machine learning methods, the team expects COSSY to meet ARPA-E’s aggressive accuracy metrics.

In order to make COSSY a secure system, all algorithms will run on local hardware, rather than in the cloud. Not only will this make the system more robust to Internet outages, but also to remote attacks.

Decreasing computational complexity will reduce the amount of needed hardware, thus lowering the cost and energy consumption of the system. This, in turn, will make COSSY more attractive to early adopters as they might recoup their investment cost within the first few years of operation.

“While we will be developing COSSY to reduce energy use in commercial buildings, there is a potential for much wider impact of this technology, from optimizing room usage in educational and office buildings to maximizing hotel revenue, both based on room-occupancy analysis,” Konrad said.