The main aim of this paper is to develop a vision-based deep learning method for real-time occupancy activity detection and recognition to help the operations of building energy systems. A faster region-based convolutional neural network was developed, trained, and deployed to an artificial intelligence (AI)-powered camera for the application of real-time occupancy activity detection. Initial experimental tests were performed within an office space of a selected case study building. The detection provided correct detections for the majority of the time (97.32%). Average detection accuracy of 92.20% was achieved for all activities. Building energy simulation of the case study building was performed to assess the potential energy savings that can be achieved. The impact of using the typical schedules and deep learning influenced profiles (DLIP) were assessed. The work has shown that the generation of the real-time DLIP from the ability to enable prediction and generation of real-time occupancy activity-based internal heat gains data can inform building energy management systems and controls of the heating, ventilation and air-conditioning (HVAC) for a more accurate and optimized operation.
Keywords Artificial intelligence, deep learning, occupancy activity, real-time detection, HVAC systems, building energy management