CO2 Sensor Occupancy Detection

A group of researchers at the University of California in Berkley have designed an algorithm that can count the number of people in a room when only the CO2 level is known.
The study, titled “Sensing by Proxy: Occupancy Detection Based on Indoor CO2 Concentration” used SenseAir’s K-30 carbon dioxide sensor module for occupancy detection. The "Sensing by Proxy" model is more accurate than previously used machine learning models, and could be used to improve the efficiency of Demand-Controlled Ventilation systems (DCV) currently in use.
DCV works on the principle that you can save energy by heating, cooling or adding fresh air to a room only when it is needed. While heating and cooling can be controlled with a thermostat, the amount of fresh air that should be added to a room can be controlled by a carbon dioxide level transmitter. When CO2 levels go up, fresh air is added until the CO2 levels return to normal (typically 10% or less of the background CO2 levels).

While CO2 transmitters in a DCV system are good at monitoring CO2 levels, they do not tell you the number of occupants in a room. This is important because indoor air quality models specify the optimum fresh air exchange rate per person. If you don’t know how many people are in a room, all the system can do is pump in fresh air until the CO2 level drops.
This is where Sensing by Proxy comes in. Using the K-30 sensor’s CO2 level data as the only input, the researchers were able to create a model “that captures the spatial and temporal features of the system and links unobserved human emission to proxy measurements of CO2 concentrations.”
The algorithm they developed was tested in both controlled and field experiments, and resulted in an error rate of 0.6 fractional persons as compared to 1.2 fractional persons by the best alternative strategy. The algorithm is also better at measuring changes, such as occupants entering and leaving a meeting. In addition, it does not require more costly Pyroelectric infrared (PIR) or Ultrasonic sensors used for current occupancy detection systems.
The algorithm is one of a growing number of papers that test the use of CO2 levels for occupancy detection. For example, Chaoyang Jiang and others in a paper titled “Indoor occupancy estimation from carbon dioxide” were able to estimate occupancy in an office with 24 cubicles. They counted occupants accurately 94 percent of the time within a tolerance of four occupants.
While DCV is initially more expensive to install, improvements like this could be used to further reduce the long-term energy costs in DCV controlled buildings.