Deep-Learning for Occupancy Detection Using Doppler Radar and Infrared Thermal Array Sensors

An accurate, real-time monitoring of the occupancy state of each space is a necessity for applications such as energy-aware smart buildings. In this paper, we have studied the feasibility of using Doppler Radar Sensors (DRS) and Infrared Thermal Array Sensors (ITA) to build an effective occupancy detection framework. The proposed sensor types are cost-effective and protect the privacy of the occupants. We have utilized Deep Neural Networks (DNN) to analyze the sensor data without any need for specialized feature extraction that is necessary for classical machine learning approaches. The results are indicative of the feasibility and the reliability of using both sensor types for detection of the occupancy state. While a threshold-based approach reached an average accuracy of 84.3% and 86% for the DRS and ITA sensors respectively, DNN models were able to achieve average accuracies of 98.9% and 99.96% for the DRS and ITA sensors respectively, thereby demonstrating the feasibility and success of the proposed framework.