Convolutional-Neural-Network-based Occupancy Behavior Pattern Recognition for Smart Elevator Dispatching

Recently there has been an increasing interest in smart elevator dispatching due to the growing pressure from high-rise buildings. Given humans are the sole users, understanding human behaviors becomes an important step towards better usage efficiency. In this regard, this paper proposes to integrate CNN-based spatial analysis of the elevator cab inside using image data from video monitoring to identify occupancy behavior patterns, and in turn to optimize the elevator dispatching to match the needs of elevator use. Occupancy behavior pattern recognition is implemented in two steps: video object detection in elevator cabs; learning and predicting usage patterns. This paper proposes to implement object detection with YOLOv3 and pattern recognition with Mask R-CNN. The proposed approach is compared with the IR-sensor-based approach through discrete event simulation for verification. A case study in an office building is reported to illustrate its feasibility and potential.