Intelligent Video Analysis Technology for Elevator Cage Abnormality Detection in Computer Vision

Abstract

The crime in the elevator cage is always a constant concern. Thus, security problem should get more attention. This paper proposes an intelligent video analysis technology for elevator cage abnormality detection in computer vision. By collecting, processing, and analyzing video images in real time, the feature vectors including the variation of foreground pixels, the variation of length and width of foreground region’s enclosing rectangle and the variation of enclosing rectangle’s center of mass are obtained. Then these feature data are processed via K-Means clustering to get observation sequences, which are used to model a Hidden Markov Model (HMMs) for the normal activity. Last, the abnormalities are identified by the log-likelihood difference from normal activity mode, and the standard value is predetermined by observing series of normal activity sequence. This paper mainly presents an overview of the technology and significant results so far achieved.