A Masked Face Detection Method for Surveillance Video

Abstract

The invention relates to a masked face detection method oriented to monitoring video. The method includes the processing strategy of video frame skipping based on simple background difference operation, the detection of moving head and shoulders region based on the improved background difference method, and the masked face detection method based on two-level Haar classifier. The processing strategy of video frame skipping based on simple background difference operation effectively narrows the processing range, reduces system overhead, and improves processing efficiency. The area containing the head and shoulders image is determined as an effective motion area, which effectively reduces the situation that the background is mistakenly detected as a moving target; combined with the frame skipping strategy, the background is only updated when a head and shoulders motion area is detected, which can further speed up the processing speed And improve the moving target extraction quality of the background difference method. According to the Haar linear features of common masked faces, the training and detection of masked objects are based on binary images, which can strengthen the gray distribution law of prominent training samples and objects to be detected, and obtain better results than simple grayscale images.

Description

Technical field [0001] The invention belongs to the field of video image processing for public safety early warning, and specifically relates to a masked face detection method. Background technique [0002] Video surveillance is widely used in the field of public safety, providing powerful data and technical support for early warning and verification in public safety management services. However, the current level of intelligent analysis for surveillance video is still relatively low, and there is no effective technical means specifically for this application in terms of masked face detection for surveillance video. Masked face detection is mainly used to quickly screen suspicious targets whose faces are deliberately occluded from massive surveillance videos. It can enhance the supervision efficiency of public security departments on specific targets, and is useful for preventing and combating crimes, tracing suspects, and maintaining social safety and stability. And so on has a...

AI Technical Summary

Problems solved by technology

[0003] For the various links in masked face detection, in terms of moving target detection, existing methods such as patent 201410110812.1 use the ViBe algorithm to establish a background model for the video image frame, and segment the foreground area by combining the ViBe algorithm of the frame difference method. The background update speed of the method is slow; patent 201110253323.8 uses edge detection and frame difference method for motion detection, and patent 201310586151.5 combines adjacent frame difference method and mixed Gaussian model to realize moving targets. The disadvantage of the above method is that it is prone to hollow areas; in human body detection On the one hand, patent 201010218630.8 uses fuzzy templates to detect multi-pose human bodies, which is slow; patent 201310415544.X uses color and depth information for human body detection methods, and the features obtained by joint feature extraction are used for human body detection, which is greatly affected by light; Patent 201110026465.0 detects people based on depth images, which is not suitable for conventional surveillance video images; in terms of masked face detection, patent 201210052716.7 detects masked heads for the entire frame of images, and the processing speed is relatively slow

Images