Face detection acceleration method based on video coding technology

Abstract

The invention discloses a face detection acceleration method based on a video coding technology. The face detection acceleration method comprises the following steps of: step A, determining a key frame (DI frame) and a reference frame (DP frame) of face detection; step B, performing face detection of a conventional process for the DI frame; step C, performing rapid face detection according to theprediction information for the DP frame, wherein the rapid face detection mainly focuses on eliminating redundant information in a face detection image and redundant steps in the face detection process; and step D, tracking a face detection result (a face rectangular frame) to provide prediction information required by the DP frame. According to the face detection acceleration method, the networkstructure of a face detection network does not need to be changed, and rapid face detection can be realized on lower-end hardware equipment.

Description

technical field [0001] The invention relates to the technical field of video coding and image pattern recognition, in particular to a face detection acceleration method based on video coding technology. Background technique [0002] Face detection is the basis for a series of applications such as face capture, face recognition, and facial reconstruction. It is widely used in security monitoring, face attendance, and smart traffic. [0003] In recent years, face detection methods based on deep learning (such as MTCNN, SSD, RetinaFace, FaceBoxes, SSH, etc.) higher requirements. How to make the face detection algorithm fast and accurate, especially the real-time application of face detection on low-end devices, has always been a concern. [0004] For the face detection acceleration method of deep learning, the simplest solution is to reduce the image size of the video frame submitted for inspection or replace the hardware device with stronger computing power, but reducing the...

AI Technical Summary

Problems solved by technology

[0004] For the face detection acceleration method of deep learning, the simplest solution is to reduce the image size of the video frame submitted for inspection or replace the hardware device with stronger computing power, but reducing the image size submitted for inspection will directly affect the minimum supported face detection size, reducing the recall rate of face detection; and upgrading hardware will bring about an increase in cost

[0005] Furthermore, in the case of the required minimum face detection size and given hardware, it is generally retrained by simplifying the network structure of the specific face detection network, and this process often requires multiple rounds of training iterations. The professional quality requirements of trainers are high, and simplifying the network structure may also lead to the loss of face detection accuracy and recall

Images