Occupancy information is important to building facility managers in terms of building operation, predictive control, safety, as well as the indoor environment quality. Previous works have addressed different occupancy counting and estimation solutions in different buildings or spaces. In this study, we adopted a single system-single zone test bed using the existing university lecture rooms to install and compare four different occupancy counting methodologies: overhead video based occupancy counting system, pantilt-zoom camera face detection system, CO2-based physical model, and CO2-based statistical model. We attempt to address the occupancy counting challenge in educational building deployment scenario with large groups of people entering and leaving. Experiments have been conducted for three months with five-minute data reporting interval. The results show that the PTZ-camera based face recognition has the most stable and highest accuracy with an R-2 of 0.972; followed by the CO2 based statistical model with an R-2 of 0.938. Discussion and improvements on the methods are discussed. A final hybrid model is proposed by using the estimated occupancy by the PTZ face detection algorithm to train the CO2 model. This plug-and-play method overcomes the practical disadvantage of these two approaches, which also overcomes the main con of all of the methods that the modeling work is required before the technologies works. (C) 2018 Elsevier B.V. All rights reserved.