An Elevator Door Blocking Behavior Recognition Method Based on Two-stage Object Detection Networks

Abstract

In recent years, the number of elevators in high-rise buildings has increasing rapidly, but the accompanying elevator safety problems become quite important and cannot be ignored. Dangerous accidents caused by bad behavior of passengers emerge in an endless stream, especially the behavior of passengers blocking elevator doors. This paper proposes a door blocking behavior recognition method based on two-stage object detection networks. It includes the Yolox-based elevator door area detection network, the EfficientNetb6-based elevator door opening degree recognition network, the final sliding average filter and logical judgments. Through the combination of deep learning algorithms for intelligent detection and intelligent monitoring system, the dangerous behavior of passengers blocking the door in the elevator monitoring video can be detected in time. The Yolox algorithm is used to detect and intercept the elevator door area in the monitoring screen, and then the feature extraction of the elevator door area is carried out through the EfficientNetb6 backbone network, and the opening value of the elevator door at each moment is obtained by regression. Finally, through a series of logical judgments, the video clips of the door blocking behavior are screened out and intercepted. The experimental results show that the method can effectively identify the door blocking behavior in the elevator, and achieve a high recognition accuracy, which is of great significance for the actual elevator monitoring scene.