Intelligent human counting through environmental sensing in closed indoor settings

Abstract

Human counting in a closed indoor environment is crucial in diverse application areas. It becomes more challenging while keeping the sensing devices hidden from eyesight. Detecting human count in such a way has significant importance in preventing any intrusion in a secure indoor space such as bank vault, treasury, armory, etc. There exist several technologies to detect human count in such closed indoor settings. Examples include surveillance based technologies and sensor based technologies. For most of these technologies, sensing devices need to be deployed in some places that remain visible to people. These happens as, if these sensing devices are not deployed in visible places, they will not be able to collect the data necessary to detect human count in the closed environments. As the sensing devices remain visible to human, there exist possibilities of being damaged by intruders. Here, intruders are able to locate the sensing devices, and therefore, they can damage these devices. As a remedy of this problem, human sensing devices could be deployed in hidden places. Thus, if these devices are made invisible to eyesight, intruders will not be able to detect locations of these devices and these devices will remain safe from being damaged. It implies that the sensing devices need to be responsive to some intangible signals, which are significantly changed due to human presence. Therefore, this study proposes a novel methodology to perform human counting in closed indoor settings based on four gaseous parameters (Carbon Dioxide, Liquefied Petroleum Gas (LPG), Nitrogen Dioxide, and Sulfur Dioxide) and two weather parameters (temperature and humidity). We conduct rigorous experiments in closed controlled environmental settings based on our proposed methodology. Besides, we leverage different machine learning algorithms such as Bagging, RandomForest, IBK, and J48 to accurately perform the counting. We achieve more than 99% accuracy for some of the classifiers in counting the number of humans present. Further, We also analyze a delicate trade-off between energy consumption and data fidelity, which will help in improving the system level performance.