A real-time video surveillance and state detection approach for elevator cabs

摘要

Manually monitoring abnormal events occurring in elevator cab is time-consuming and hard to response in time. In this paper, we develop an intelligent video surveillance and state identification system for elevator cabs, in which the information such as the state of cab door, the behavior of human body, as well as the number of people in an elevator cab, can be estimated and evaluated for security. In our proposed framework, the state detection of the elevator door is first implemented by an improved ViBe algorithm; afterwards, a trained deep learning model, i.e., an SSD_MobileNet model, is trained to accurately estimate the number of people in the elevator cab; finally, a state identification approach in the case of two abnormal behaviors occurring in cabs is developed through the ViBe and optical flow based algorithms, respectively. To evaluate the effectiveness of our proposed system, several real surveillance videos from elevator cabs are tested, and the results show that the accuracy of door detection is reached by 95.6%, the accuracy of counting people in cabs by 94.1%, and the accuracy of abnormal events detection in cabs by 92.0% on average.