A high-precision fast face detection method

Abstract

The invention discloses a high-precision fast face detection method, which relates to the technical field of face recognition. The method guarantees the detection speed and improves the accuracy of face recognition by designing lightweight network structures such as fast digestion convolution layer and multi-scale convolution layer, and combining with a new anchor point densification strategy. Atthat same time, the dynamic scale cross entropy is obtained by combining the modulation factor with the standard cross entropy loss remodeling, so as to construct a brand-new loss function, In order to solve the problem of extreme foreground background imbalance, while ensuring the detection speed, the facial recognition accuracy is further improved. Ultimately, it can not only run on a single CPUat the speed of 65FPS, but also significantly improve the recall rate of the small face, which fully meets the needs of the project.

Description

technical field [0001] The invention relates to the technical field of face recognition, in particular to a high-precision fast face detection method. Background technique [0002] With the continuous development of the field of face detection, modern face detection methods can be roughly divided into two categories: one is based on manual detection, and the other is based on CNN (Convolutional Neural Network, Convolutional Neural Network) detection. [0003] In recent years, CNN-based face detectors have developed rapidly: Farfade uses enhanced functions on top of CNN functions for face detection; Faceness trains a series of CNNs for facial attribute recognition and can detect under partial occlusion; CascadeCNN developed a CNN-based cascade architecture with strong discriminative ability and high performance; UnitBox introduced a new cross-joint loss function; CMS-RCNN used faster RCNN and context information in face detection; Convnet combined CNN It is successfully inte...

AI Technical Summary

Problems solved by technology

Although face detection has made great progress, there are still huge challenges in practical engineering. For example, it is difficult to achieve real-time speed and maintain high performance on the CPU.

Images