Cascaded convolutional neural network based human face occlusion detection method

Abstract

The invention discloses a cascaded convolutional neural network based human face occlusion detection method. The method comprises the following steps of 1) obtaining a video frame image; 2) performing normalization processing on the image, and copying and storing two images; 3) graying the image 1 and performing histogram equalization processing on the brightness imbalance image; 4) performing human head detection in a multi-scale sliding window form by adopting a three-level cascaded network, and storing window coordinates and sizes which meet the conditions; 5) performing clustering analysis on the window coordinates to obtain a target window position; 6) according to the obtained data, intercepting a human head region in the image 2, and performing normalization processing and brightness adjustment; and 7) performing eye and mouth detection in the multi-scale sliding window form by adopting a two-level eye / mouth cascaded network regionally, and if a set condition is not met, judging that the eyes / the mouth is occluded and triggering alarming. The method is strong in illumination and posture robustness, suitable for various occlusion types, and relatively high in detection precision.

Description

technical field [0001] The present invention relates to neural network, computer vision, image processing, pattern recognition and other technical fields, especially a method for processing, analyzing and understanding video signals, controlling a monitoring system, and realizing face occlusion detection. The method can be used In public places such as schools, banks, prisons, factories, etc., it is also applicable to the access control and surrounding areas of private houses. Background technique [0002] Biometric identification technology is a technology that uses automated technology to detect personal physiological characteristics or personal behavior characteristics for identity verification. It has been widely used in the fields of business, military, and criminal investigation. Among the many biometric features, face recognition technology has important academic research value and broad application prospects due to its advantages of initiative, non-invasiveness, user...

AI Technical Summary

Problems solved by technology

As the proportion of people wearing glasses in daily life increases, the problems caused by glasses blocking are becoming more and more common; environmental pollution has caused more and more common wearing masks when going out; in practical applications, such as smart access control, video Surveillance, security systems, criminal identification, etc., basically collect face images in a non-cooperative environment, which is easily blocked by other people or objects

[0003] These interference factors that cause occlusion make the face data acquired by the imaging device incomplete, causing the loss of part or even all of the face information, resulting in the inability of the face recognition system to extract complete and effective face information, which affects the entire face authentication system. The accuracy of detection and identification

Due to the variety of occlusion types, random positions, and uncertain sizes, there is no suitable method to model occlusion, which makes it very difficult to deal with occlusion problems

Images