Research On The Masked Face Discrimination Method On ATM Machines

With the development of information technology and the emergence of ATM machines,people gradually use ATM machines to take out services such as withdrawals,deposits,and transfers instead of bank counter services.The emergence of ATMs has brought a lot of convenience to people for banking services,and it has been widely used,but at the same time,some criminals who use their opportunistic methods to conduct crimes such as money laundering,stealing other people's property,and other crimes by hiding their faces are illegal activities.There is a hidden danger in the security of bank users' assets and the financial turnover of banks.Therefore,there is an urgent need to research an efficient masked face discrimination method on ATM Machine for use in intelligent video surveillance systems to prevent the occurrence of such cases.Aiming at a series of criminal activities by law-breaking criminals on ATM machines by covering the face,this paper proposes a method for the masked face discrimination on ATM machines,which can quickly discriminate the masked face on ATM machines.The method is mainly divided into two steps:In the first step,the strategy of using the YOLO model to quickly detect the target face is proposed,and the YOLO model is retrained by collecting related face data sets.The face of the target area is quickly returned and the entire image is directly used as the network.The input directly returns to the face position at the output end,completing the rapid regression from the input of the original image to the position of any face.Compared to other literature-related methods,this paper uses the YOLO model to locate the face,with higher detection accuracy and faster detection rate,and achieved the purpose of quickly locating the face,thereby excluding the interference of the background scene to the next occlusion discrimination.In the second step,the OLIB-based multi-angle facial feature point detection method is used to identify the occluded face,and the extracted HOG feature is used to perform cascade regression on the feature points of the face.The facial features can be detected according to whether the face can be detected.The 68 feature points are used for further blocking determination.If it can be matched to the corresponding 68 feature points of the face,it is determined as a normal face,and if not,it is determined as masked face and then prohibited from further operating,thus ensures the accuracy of the masked face recognition and the proposed method can adapt to the occlusion discrimination of various situations,and the discrimination speed rate is faster than the traditional methods.By comparing the masked detection rate,missed detection rate,false detection rate and detection speed rate of the relevant methods by experiments,it is verified that the method is more effective.It can detect all kinds of occlusion accurately and quickly,and the overall performance is excellent,and realized the requirement of real-time and robustness for the masked face discrimination on the ATM,and has a high application value.