Jönköping AI Lab (JAIL), Department of Computer Science and Informatics, School of Engineering, Jönköping University, 551 11 Jönköping, Sweden.
Sensors (Basel). 2021 Feb 3;21(4):1036. doi: 10.3390/s21041036.
Indoor occupancy prediction is a prerequisite for the management of energy consumption, security, health, and other systems in smart buildings. Previous studies have shown that buildings that automatize their heating, lighting, air conditioning, and ventilation systems through considering the occupancy and activity information might reduce energy consumption by more than 50%. However, it is difficult to use high-resolution sensors and cameras for occupancy prediction due to privacy concerns. In this paper, we propose a novel solution for predicting occupancy using multiple low-cost and low-resolution heat sensors. We suggest two different methods for fusing and processing the data captured from multiple heat sensors and we use a Convolutional Neural Network for predicting occupancy. We conduct experiments to assess both the performance of the proposed solutions and analyze the impact of sensor field view overlaps on the prediction results. In summary, our experimental results show that the implemented solutions show high occupancy prediction accuracy and real-time processing capabilities.
室内人员密度预测是智能建筑中能源消耗、安全、健康和其他系统管理的前提。以往的研究表明，通过考虑人员密度和活动信息来自动化建筑物的供暖、照明、空调和通风系统，可以将能源消耗降低 50%以上。然而，由于隐私问题，很难使用高分辨率传感器和摄像头进行人员密度预测。在本文中，我们提出了一种使用多个低成本、低分辨率热传感器进行人员密度预测的新方法。我们提出了两种不同的方法来融合和处理从多个热传感器捕获的数据，并使用卷积神经网络来预测人员密度。我们进行了实验来评估所提出的解决方案的性能，并分析了传感器视场重叠对预测结果的影响。总的来说，我们的实验结果表明，所实现的解决方案具有较高的人员密度预测精度和实时处理能力。
