Anchor-point-free face detection method and system

Abstract

The invention discloses an anchor-point-free face detection method and a system. The method comprises the steps of inputting a to-be-detected face image into a trained face detection network for facedetection; performing face feature extraction on the trained face detection network, and outputting a face heat map, a face scale map and a face center offset map; regarding points greater than a preset threshold in the human face heat map as human faces; and extracting a face coordinate offset from a corresponding position on the face center offset map, adding the face coordinate offset to the coordinates of the face heat map to obtain a final face center position, and finally calculating the width and height of the face on the face scale map to obtain the coordinates of the face. According to the method, the human face is expressed as the central point of the human face frame, and then the size of the human face frame is directly returned according to the image features of the central position, so that the post-processing time of tedious anchor points is reduced, and a rapid and efficient human face detection task is realized.

Description

technical field [0001] The invention relates to the field of face detection based on deep learning, in particular to an anchor-free face detection method and system. Background technique [0002] Face detection is one of the fundamental problems in computer vision and pattern recognition, widely used in mobile devices and embedded devices. Because, these devices usually have limited memory storage and low computing power. Therefore, it is necessary to quickly and accurately detect faces. [0003] With the great breakthrough of convolutional neural network, face detection technology has made remarkable progress in recent years. Early on, face detection algorithms with cascaded convolutional neural network frameworks used cascaded networks to learn face features to improve performance and maintain efficiency. However, there are some problems with cascaded convolutional neural network detectors: 1) The running time of the detector is negatively correlated with the number of ...

AI Technical Summary

Problems solved by technology

However, there are some problems with cascaded convolutional neural network detectors: 1) The running time of the detector is negatively correlated with the number of faces on the input image; 2) Since these methods optimize each module separately, the training process is extremely complicated

However, these methods have some shortcomings. On the one hand, a large number of dense anchor points are usually required to obtain a good recall rate, which leads to a serious time-consuming post-processing process.

On the other hand, the anchor point is a hyperparameter design based on statistical calculations for a specific dataset, and its generalization ability is poor

[0005] In addition, the current state-of-the-art face detection technology usually uses a larger backbone network such as VGG16, Resnet50 / 152, which makes it difficult to use in practical applications because the network consumes too much time

Images