Research On Abnormal Behavior Recognition In The Elevator

Intelligent video surveillance technology has been widely used in theexisting video surveillance scenes, image processing and video analysistechnology are applied to the elevator video surveillance scene.To the particularenvironment of the elevator,the potential abnormal issues are divided into threecategories.The corresponding abnormal behavior detection methods are proposedrespectively according to the characteristics of different behavior.To the particular environment in the elevator, the potential abnormal issuescould be divided into three categories in this article: fainted behavior of singlepeople, people fighting, looting and dangerous legacy. According to thecharacteristics when the above three kinds of behavior occurred, threecorresponding anomaly detection algorithms are proposed in this article.Firstly,the adaptive median filtering method and the average method are fused toobtained the background model and the background reduction method combinedwith edge detection method of Sobel operator are used to extract the foregroundimage.The inadaptability problem in elevator surveillance scenes of the existingforeground extraction algorithms has been solved by this method. The effectiveprospects are obtained while discarding the useless edges, so as to get thecomplete prospect target. On faint of single people recognition problem, it couldbe determined whether there are static prospects by calculating the Hausdorffdistance of the contours between two adjacent frames.The size of the data and thecomplexity of computation are greatly reduced by the improved pyramid Lucas-Kanade algorithm based on Harris corner detection method when calculating theoptical flow vector. This paper established a model based on kinetic energy ofcorner as a decision criteria of abnormal behavior recognition. The backgroundsubtraction method and the frame difference method are used in the detectionalgorithm of legacy, when background subtraction method could detect objects while frame difference method could not detect them,it could be determined thatthere are some left objects in the elevator.The system platform has been developed based on the MFC language andusing OpenCV vision libraries and there are a number of related experimentshave been carried on in this simulated environment. The experimental resultsshow that the algorithm could detect most of the abnormal behavior occurred inthe elevator and meet the demand of elevator video surveillance system with highaccuracy and could alarm real-time.