Face Occlusion Detection Method Based on Deep Convolutional Neural Network

Abstract

The invention discloses a face occlusion detection method based on a deep convolutional neural network, comprising: dividing an input image into blocks to obtain a target preselected area; The first deep convolutional neural network of the network and the first multi-layer perceptron connected to it obtains the required parameters, extracts the features of the target pre-selected area and classifies them; according to the extracted features, predicts the position of the head by the second multi-layer perceptron; The classification category is the credibility of the head and the predicted head position through non-maximum value suppression filtering to remove overlapping repeated detection frames; combined with the original image segmentation to obtain the head block, and construct the second deep convolutional neural network based on the multi-task learning strategy Network to determine whether the left eye, right eye, nose and mouth of the person's head block are blocked. This method can accurately detect blocked faces and judge their specific blocked parts, and is mainly used for crime warning of the video camera in front of the ATM.

Description

technical field [0001] The present invention relates to a face occlusion detection method, in particular to a face occlusion detection method based on a deep convolutional neural network. Background technique [0002] Automatic Teller Machines (ATMs) have been a target of criminals since their widespread introduction in the 1970s. For example, scammers use various means to obtain user card numbers and passwords. Real-time automatic alarm system is the most direct technology to solve this problem. Because surveillance cameras are installed on almost all ATMs. However, the video needs to be supervised by humans 24 hours a day, but human fatigue and distraction will be inevitable. Therefore, politicians and businessmen urgently need a face occlusion detection method for ATMs. [0003] Face occlusion detection has been studied for several years, and several methods have been proposed, many of which are aimed at enhancing the security of ATMs. However, its feature expression...

AI Technical Summary

Problems solved by technology

[0004] Recently, the most successful method of object detection is to use the well-known sliding window (sliding window) mode. However, in order to accurately detect objects with large size changes, this method will lead to a sharp increase in the calculation of subsequent classifiers.

Images