Paper Conference

Proceedings of eSim 2016: 9th Conference of IBPSA-Canada


Detecting occupants’ presence in office spaces: a case study

H. Burak Gunay, Anthony Fuller, William O’Brien, Ian Beausoleil-Morrison

Abstract: This paper introduces and assesses the predictive accuracy of five occupancy detection methods using data gathered through passive-infrared (PIR), CO₂, door contact sensors, and a camera in an intermittently used shared-office space. Recommendations for PIR sensor placement and optimal delay time choices were developed. A recursive data-driven modelling methodology was introduced to filter out occupants’ presence from the CO₂ concentration response of an office space. The potential of employing an image recognition-based presence detection algorithm was investigated. The simplest way of tracking occupancy in shared-office spaces – where a single PIR cannot have a direct line-of-sight for all occupants – was found to place a sensor facing the door with an adaptive delay period selecting algorithm. When available, use of additional sensing such as a network of PIRs, door contact, and CO₂ sensors can improve the detection accuracy of intermediate vacancy periods. List of abbreviations AHU Air-handling unit BAS Building automation system BPS Building performance simulation HVAC Heating, ventilation, and airconditioning NDIR Nondispersive infrared sensor PIR Passive-infrared RFID Radio frequency identification VAV Variable air volume List of symbols 𝑝 Probability p Presence (0 or 1) ∆𝑡 Time between two consecutive movement detections 𝐶i𝑛 Indoor CO₂ concentration (ppm) 𝐶𝑚 Measured CO₂ concentration (ppm) 𝐶𝑜𝑢𝑡 Outdoor CO₂ concentration (ppm) 𝐶𝑠𝑎 Supply air CO₂ concentration (ppm) and Robinson 2011), these occupancy models need to represent intermediate occupancy and vacancy periods realistically. For example, simulated occupants in most light switch, window blinds and operable window use models are set to undertake actions more frequently at arrival or departure timesteps; and the occupancy models’ ability to predict the timing and frequency of the intermediate arrival or departure events is intrinsically dependent on the quality of the occupancy observations used in developing these models. However, the data used in developing occupancy models often were not assessed against the ground-truth. In addition, detecting occupants’ presence in office spaces has become essential for many lighting and HVAC control applications − e.g., demand controlled ventilation, motion sensor-based lighting control (Agarwal et al. 2010). Particularly, incorrect vacancy detections for lighting controls can lead to occupant annoyance, and in many cases occupants place an opaque material to cover the surface of motion detectors (O'Brien and Gunay 2014). In addition, there has been a surge of research on innovative lighting and HVAC controls which simply takes the detection of human presence in office spaces for granted – e.g., Gunay et al. (2015)’s occupancy learning adaptive temperature setback algorithm and Roisin et al. (2008)’s lighting control algorithm. The effectiveness of such control strategies is strictly dependent on the quality of presence detections. Background PIR sensors are the most common sensor type used in office buildings to detect human presence. They detect movements from changes in the infraredradiation impinging on them (Dodier et al. 2006). Given that movements are discrete events, in practice a delay value (e.g., 15 to 60 min) is heuristically selected to avoid incorrect vacancy detections during immobility. After each movement detection, the space is assumed occupied for this delay period. The uncertainties in occupants’ activeness (frequency of 𝑛𝑜𝑐𝑐 Number of occupants detectable movements), office layouts and sensor positioning play an unprecedented role over the 𝑝𝑠𝑎 Pressure of supply air (Pa)
Pages: 185 - 195