Indoor occupancy measurement by the fusion of motion detection and static estimation

Abstract

The human dimension information in buildings is crucial for energy saving, environment conditioning, human health, and security management. Indoor occupancy information measurement methods based on cameras mainly consist of cameras at the room entrance and interior. However, each method has particular advantages and limitations. In this paper, we propose a four-step occupancy detection and estimation system based on cameras. Firstly, we develop an occupancy detection method to filter the non-occupied frames and provide the switching value. Secondly, we design a motion detection method to detect entering or leaving events at the room entrance. Thirdly, Fully Convolutional Head Detector (FCHD) is used to detect indoor human heads. Finally, considering the occupancy number information depends on the previous results, we propose a fusion estimation method that combines motion detection and static estimation by calculating the Kalman filter and Occupancy Frequency Histogram (OFH). The novel fusion method takes the advantages of two methods and overcomes its own limitations, so that it can compensate each other to reach the high-accuracy results. The accumulative error and the shaking results are cleared, and then some headlike objects are filtered. The proposed fusion method reaches an occupancy detection accuracy of 97.8%; reaches an occupancy estimation Score of 78.52% and 79.18%; reaches an occupancy estimation the Mean Square Error (MSE) of 0.21 and 0.23. The proposed method is novel and effective for fusing these two vision situations through practical experimental verification. The proposed method has a significant potential to save energy and prevent the spreading of coronavirus disease 2019 (COVID-19) in buildings.