Multi-face real-time tracking and preferential algorithm

Abstract

The invention relates to the technical field of communication, in particular to a multi-face real-time tracking and preferential algorithm, which comprises the following steps of: obtaining an image which needs to be processed currently in a video stream to obtain a first frame of face image; performing detection and quality evaluation on the obtained first frame of face image to obtain coordinatevalues and quality evaluation values of all face frames of the current frame; judging whether the face quality evaluation score meets the requirement or not; replacing the coordinate values and the quality evaluation values of all the face frames of the first frame with the coordinate values and the quality evaluation values of all the face frames of the current frame; and traversing and calculating the IOU values of the face frames in the coordinate values and the quality evaluation values of all the face frames of the current frame. According to the invention, IOU value calculation is carried out on the face frames detected by the front and rear frames in the video stream to determine whether the face frames detected by the front and rear frames are the same person, and if yes, the sameperson determines whether to carry out face recognition in combination with the quality evaluation value of the face picture, thereby effectively reducing unnecessary repeated uploading and recognition of the same face picture.

Description

technical field [0001] The invention relates to the field of communication technology, in particular to a multi-face real-time tracking and optimization algorithm. Background technique [0002] With the wide application of deep learning in computer vision, face recognition has been implemented in many projects. Face recognition includes face detection, face selection, key point positioning, face alignment, face feature value extraction and comparison. , quickly and effectively screen high-quality face pictures of the same person at different times under unconstrained (non-sense, non-cooperative) conditions and use a few of them for face recognition in real time, which can greatly reduce unnecessary Repeated recognition, especially when doing face detection on the front end and then uploading face pictures to the back end for face recognition, also greatly reduces unnecessary repeated uploads; [0003] Judging whether they are the same person based on the intersection ratio ...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to solve the phenomenon of face loss existing in the prior art, resulting in too few service pushes and repeated uploading and recognition of face recognition in actual scenarios, and proposes a multi-face real-time Tracking and Merit Algorithms

Images