Tiny face detection method, system and device based on label box virtual amplification

Abstract

The invention belongs to the field of visual detection and image processing, particularly relates to a micro face detection method, system and device based on label box virtual amplification, and aimsto solve the problems that only a small number of micro face feature map areas are defined to be positive in model training, so that the detection precision and accuracy of a model on micro faces arerelatively low. The method comprises the following steps: marking a tiny face by using a rectangular frame; constructing a face detection model based on LFFD; based on the down-sampling multiple of the feature map of the model for the small-scale face detection branches, performing virtual amplification of a rectangular frame; generating classification and prediction targets of the model based onthe rectangular frame and the virtual amplification frame respectively; performing model training through back propagation and random gradient descent method iteration; and carrying out online face detection through a model for simultaneously detecting a face with a general size and a tiny face after training. According to the invention, the label box virtual amplification is applied to the microface data, the detection rate of the micro face is improved, and the micro face detection is achieved.

Description

technical field [0001] The invention belongs to the field of visual detection and image processing, and in particular relates to a tiny human face detection method, system and equipment based on virtual amplification of a label frame. Background technique [0002] In the field of security monitoring, face detection algorithm, as an extremely important technology in face recognition, has been widely used. At this stage, the face detection algorithm faces various challenges brought about by the open monitoring environment, such as backlight, insufficient light, partial occlusion, abnormal posture, etc. Additionally, face detection algorithms are typically designed to detect larger faces (typically greater than 40 pixels), i.e., closer to the camera. If you want to detect long-distance faces without upgrading the camera, you need to introduce tiny face detection technology. Tiny face detection not only requires the ability to detect faces of normal size, but also needs to det...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the above-mentioned problems in the prior art, that is, only a small number of small face feature map areas are defined as positive in model training, which leads to the problem that the detection accuracy and accuracy of the model for small faces are low, the present invention provides A tiny face detection method based on the virtual augmentation of the annotation frame is proposed. The detection method includes the following steps:

Images