The Research On Abnormal Behavior Identification Of The ATM User

With the in-depth development of e-finance and the continuous advance of the goldencard project, the number of ATM is increasing and it is widely used while there are manycriminal activities about using ATM machines to steal the money of the legitimate cardholderand malignant events about the destruction of ATM. However, the lack of intelligentmonitoring methods in the monitoring system of the bank causes the criminal rates rising yearby year. Therefore, it is an urgent problem to find out exceptional situation timely to avoid thehappen of crime events, to protect the securely using of the bank ATM, to protect the interestsof the legitimate cardholder and the bank and to guard against all kinds of criminal behavior.According to practical problems which occur in the process of using ATM machines, thispaper sets deposit video of a real ATM machine user as input sequence, analysis and study thebehavior of user in order to detect the abnormal behavior of the user in the ATM machinescene. The main work and research results of this paper can be summarized as follows:In this paper, the adaptive gaussian background model algorithm and the face detectionalgorithm which based on elliptic skin model are used to detect the human body and face inthe ATM machine scene. We detailed discuss the Hu moments algorithm and the theory thatthe improved Hu moments algorithm can extract the feature of the moving target in the caseof meeting the translation, scaling, rotating invariance in discrete and continuous state. Inaddition, we study the advantages and disadvantages of existing tracking algorithm. Then，thispaper adopts mean-shift tracking algorithm which has the characteristics of simple calculationand good real-time to track moving objects in the monitoring area.This paper proposes two methods of detecting and identifying abnormal behaviors forthe abnormal behavior of user in the ATM machine scene. First, according to the condition ofwhether appear camouflage withdrawals with the masked face or not when users operate ATMmachine, we use detection algorithm which based on the YCbCr elliptical model to detectregional face and use predefined semantic abnormal to judge abnormal behavior of the target.Second, we set the whole movement of objects in ATM airport as the research object, train acontinuous hidden markov model with the left-right structure and experiment simulationabout the normal behavior of withdrawals which is the moving object in the real ATM airportscene. We judge whether the behavior of moving object belongs to abnormal behavior or notby calculating the output-probability of testing samples to the trained models.