The Abnormal Behavior Detection Of ATM Operation On Computer Vision

Extensive use of Automatic Teller Machine (ATM) facilitates a number ofdepositors to a large extent and ATM has become an important tool of convenientliving. But in recent years, the endless stream of ATM crimes has a serious impact onmany people’s property. Currently, the security measure is installing the surveillancecameras near the ATM, which requires a large number of security personnel around theclock to view surveillance video to check whether there is a crime, or to look atsurveillance video to get clues in the aftermath of crime. This approach has problemsof spending too many manpower, material resources and being low real-time.For the real-time direction to the abnormal behavior of ATM, this paper designedintelligence video surveillance system of ATM. Aiming at the video image collectedwithin the region, extract human features, analyze from multiple angles and thendetermine whether there is abnormal behavior depending on the data coming from thecomprehensive analysis of four modules. In the face detection module, using Adaboostalgorithm, extract the face in the body and issue the alerts for the phenomenon of theface block. In strenuous exercise detection module, this paper proposed the violentmotion detection algorithm based on mass variability and centered variability of bodyimage. Due to the demand of this article and better applicability, according to thecomprehensive probability function of these two variability, this algorithm coulddetermine whether there is strenuous exercise, has good real-time and high recognitionrate. In the hand trajectory recognition module, according to the system characteristics,this passage improves the skin detection algorithm based on YCbCr color space,narrowing the scope of testing and detecting effectively the hand movement; thenusing Mean Shift Algorithm, track the detected hand movement and extract the handtrajectory; model the hand trajectory using Hidden Markov Model; finally do the tracematching. In the module of abandoned objects detection, this passage using thecombined algorithms of Temporal Difference and Background Subtraction, accordingto comprehensive data analysis of two simple algorithms results, could effectivelydistinguish moving object, noises, background images and stationary objects, enhancethe real-time of remnants detection and reduce the identification error. Finally, thispassage develops the intelligence video surveillance system for ATM based onOpenCV, and describes the implementation of each module.Our work is imbursed by the Key Technologies R&D Program of Shaanxi (GrantNO.2009K08-11).