CN113869210A

CN113869210A - Face recognition following method and intelligent device adopting same

Info

Publication number: CN113869210A
Application number: CN202111144329.1A
Authority: CN
Inventors: 梁贵轩; 马焱坤; 周可佳
Original assignee: China Comservice Enrising Information Technology Co Ltd
Current assignee: China Comservice Enrising Information Technology Co Ltd
Priority date: 2021-09-28
Filing date: 2021-09-28
Publication date: 2021-12-31

Abstract

The invention discloses a face recognition following method and an intelligent device adopting the method, wherein the method comprises the following steps: carrying out face recognition on the face in the visual field; and tracking the face after face recognition. The principle of the invention is as follows: the person after face recognition still exists in the following frames at a high probability, so that the purpose of following is achieved by tracking the face without carrying out recognition work with high system overhead on each frame.

Description

Face recognition following method and intelligent device adopting same

Technical Field

The invention relates to a face recognition following method and intelligent equipment carrying the face recognition method.

Background

Face recognition is a biometric technology for identity recognition based on facial feature information of a person. A series of related technologies, also commonly called face recognition and face recognition, are used to collect images or video streams containing faces by using a camera or a video camera, automatically detect and track the faces in the images, and then perform face recognition on the detected faces.

For intelligent devices, especially intelligent robots, the purpose of face recognition is not only identification, but also follow-up after identification. Most of the existing face recognition algorithms use Dilb to perform real-time face recognition, that is, face recognition is performed once for each frame of video. Although this algorithm is guaranteed in terms of accuracy, the overhead of repeated recognition is extremely large, resulting in seizure.

Disclosure of Invention

In view of the above, the present invention provides a face recognition method and an intelligent device using the face recognition method, which change real-time face recognition into a face recognition and then tracking manner, thereby greatly reducing system overhead and improving system fluency.

In order to solve the above technical problems, the present invention provides a face recognition following method, comprising: carrying out face recognition on the face in the visual field; and tracking the face after face recognition. The principle of the invention is as follows: the person after face recognition still exists in the following frames at a high probability, so that the purpose of following is achieved by tracking the face without carrying out recognition work with high system overhead on each frame.

As an improvement, the face recognition is carried out in an initial frame, and tracking is carried out in a subsequent frame. In the following activity of recognizing a certain face, only the first frame is subjected to face recognition to determine the identity, and the rest of the subsequent frames are only subjected to tracking work, so that the system overhead is reduced to the maximum extent.

As an improvement, the face recognition comprises: calculating a 128D characteristic value of each photo face according to a plurality of pictures of the same face, and calculating a 128D characteristic mean value of all photo faces as a reference value; and calculating a 128D characteristic value of the face picture acquired in real time as a comparison value, comparing the comparison value with a reference value, judging as the same person if the difference value is smaller than a threshold value, and judging as a stranger if the difference value is larger than the threshold value.

Preferably, the calculating the 128D feature value of each photo face according to a plurality of pictures of the same face comprises: positioning a face in the collected video, drawing a face frame by taking the face as a center, and storing an image in the face frame; the 128D feature values of the faces in each image are calculated.

As an improvement, the 128D characteristic value of each photo face is calculated through a Dlib resnet50 depth residual neural network.

Preferably, the euclidean distance between the comparison value and the reference value is calculated by comparing the comparison value with the reference value.

As an improvement, the tracking according to the face recognition result includes: setting and registering feature frames for all human faces in each frame of image in the video, and calculating the mass center of each feature frame; sequentially calculating Euclidean distances of all particles in adjacent frames, and judging that two particles with the shortest Euclidean distance between the adjacent frames represent the same face.

As an improvement, when a new face appears in the tracking process, setting a feature frame for the new face, registering, and calculating feature frame particles; sequentially calculating Euclidean distances of all particles in adjacent frames from the frame, and judging the two particles with the shortest Euclidean distance between the adjacent frames as representing the same face; and judging the face represented by the particles which are far away from the particles in the previous frame as a new face, and identifying and tracking the new face again.

As an improvement, when the number of frames of any face disappearing in the tracking process exceeds a threshold value, the face is cancelled.

The invention also provides intelligent equipment carrying the face recognition following method.

The invention has the advantages that: the face recognition following method with the steps replaces the existing real-time recognition by a recognition and tracking mode, reduces the recognition process which consumes most resources, and greatly increases the system fluency on the premise of not reducing the accuracy basically.

Drawings

FIG. 1 is a flow chart of the present invention.

Detailed Description

In order that those skilled in the art will better understand the technical solutions of the present invention, the present invention will be further described in detail with reference to the following embodiments.

As shown in fig. 1, the present invention provides a face recognition following method, which comprises the following steps:

s1, carrying out face recognition on the face in the visual field; face recognition is performed at the initial frame.

And S2, tracking the face by the following frames after face recognition.

For example, two persons exist in the Nth frame, and Zhang III and Lile IV are known through traversal comparison; then for the subsequent frames (N +1, N +2, etc.) in the video stream, if there are two people, it can be determined approximately whether they are Zhang three or Li four, only the position in the frame has changed (this is also a precondition that object tracking can be applied); therefore, only the corresponding relationship between two persons in the frame N and the two persons in the frame N +1 needs to be determined, and the feature descriptor extraction for the frame N +1 does not need to be performed again, and then traversal comparison is performed, so that a large amount of system resources are consumed.

The face recognition in step S1 specifically includes the following steps:

s11, calculating a 128D characteristic value of each photo face according to a plurality of pictures of the same face, and calculating a 128D characteristic mean value of all photo faces as a reference value;

s12, a 128D characteristic value is calculated for the face picture collected in real time to serve as a comparison value, the comparison value is compared with a reference value, if the difference value is smaller than a threshold value, the same person is judged, and if the difference value is larger than the threshold value, a stranger is judged. Specifically, the euclidean distance between the comparison value and the reference value is calculated.

Step S11 specifically includes:

s111, positioning the face in the collected video by using a Dilb algorithm, drawing a face frame by taking the face as a center, preferably 480 × 640 in the embodiment, and storing the image in the face frame;

s112 calculates a 128D feature value of the face in each image. In this embodiment, the 128D feature value of each photo face is calculated by a Dlib resnet50 depth residual neural network.

The step S2 of tracking according to the face recognition result specifically includes:

s21, setting and registering feature frames for all faces in each frame of image in the video, and calculating the mass center of each feature frame;

s22 sequentially calculates the euclidean distances of all particles in adjacent frames, and determines that two particles having the shortest euclidean distance between adjacent frames represent the same face.

If two faces three and four are recognized in the N frame, three faces and four faces are registered in the system in the N +1 frame for continuous tracking, a feature frame is set for each face, and mass points, P, of each feature frame are calculated₁₁And P₁₂The two particles represent Zhang three and Li four, respectively. Similarly, a feature frame is set for each face in the N +2 frames, and the mass point P of each feature frame is calculated₂₁And P₂₂. Respectively calculate P₁₁To P₂₁And P₂₂Distance and P₁₂To P₂₁And P₂₂The distance of (c). If P is₁₁To P₂₁Is less than P₁₁To P₂₂Is determined, then P is determined₂₁Zhang III. In the same way, if P₁₂To P₂₂Is less than P₁₂To P₂₁Is determined, then P is determined₂₂Is Liqu. And so on for N + N frames.

When a new face appears in the tracking process, setting a feature frame for the new face, registering, and calculating feature frame particles; and sequentially calculating Euclidean distances of all particles in the adjacent frames from the frame, and judging that the two particles with the shortest Euclidean distance between the adjacent frames represent the same face. And judging the face represented by the particles which are far away from the particles in the previous frame as a new face, and identifying and tracking the new face again. For example, Wang Wu appears in the field of view at the time of N +3 frames, and three particles P appear₃₁、P₃₂、P₃₃At this time, P is required to be calculated separately₂₁And P₂₂To P₃₁、P₃₂、P₃₃Of Euclidean distance between, from, P₂₁And P₂₂And the remote point is a new face, which needs to be subjected to face recognition again, and the face is continuously tracked according to the tracking method after the face is recognized.

In addition, when the number of the lost frames of any face exceeds the threshold value in the tracking process, the face is cancelled. For example, particle P representing Liqu at frame N +5₅₂By exceeding a set number of frames from view (depending on the application scenario)Instead, the set frame number may be 1 to n), then lee is injected from the system and not tracked.

The invention also provides intelligent equipment carrying the face recognition following method, and the intelligent equipment can realize following transportation, protection and other functions by the face following method.

The above is only a preferred embodiment of the present invention, and it should be noted that the above preferred embodiment should not be considered as limiting the present invention, and the protection scope of the present invention should be subject to the scope defined by the claims. It will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the spirit and scope of the invention, and these modifications and adaptations should be considered within the scope of the invention.

Claims

1. A face recognition following method is characterized by comprising the following steps:

carrying out face recognition on the face in the visual field;

and tracking the face after face recognition.

2. The face recognition following method according to claim 1, wherein:

the face recognition is carried out in the initial frame, and the tracking is carried out in the subsequent frame.

3. The face recognition following method according to claim 1, wherein the face recognition comprises:

calculating a 128D characteristic value of each photo face according to a plurality of pictures of the same face, and calculating a 128D characteristic mean value of all photo faces as a reference value;

and calculating a 128D characteristic value of the face picture acquired in real time as a comparison value, comparing the comparison value with a reference value, judging as the same person if the difference value is smaller than a threshold value, and judging as a stranger if the difference value is larger than the threshold value.

4. The face recognition following method according to claim 3, wherein the calculating the 128D feature value of each photo face from a plurality of pictures of the same face comprises:

positioning a face in the collected video, drawing a face frame by taking the face as a center, and storing an image in the face frame;

the 128D feature values of the faces in each image are calculated.

5. The face recognition following method according to claim 3, wherein the 128D characteristic value of each photo face is calculated by a Dlib resnet50 depth residual neural network.

6. The face recognition following method according to claim 3, wherein the comparing the comparison value with the reference value calculates the Euclidean distance between the comparison value and the reference value.

7. The face recognition following method according to claim 1, wherein the tracking according to the face recognition result comprises:

setting and registering feature frames for all human faces in each frame of image in the video, and calculating the mass center of each feature frame;

sequentially calculating Euclidean distances of all particles in adjacent frames, and judging that two particles with the shortest Euclidean distance between the adjacent frames represent the same face.

8. The face recognition following method according to claim 7, wherein: when a new face appears in the tracking process, setting a feature frame for the new face, registering, and calculating feature frame particles; sequentially calculating Euclidean distances of all particles in adjacent frames from the frame, and judging the two particles with the shortest Euclidean distance between the adjacent frames as representing the same face; and judging the face represented by the particles which are far away from the particles in the previous frame as a new face, and identifying and tracking the new face again.

9. The face recognition following method according to claim 7, wherein: and when the number of the lost frames of any face exceeds a threshold value in the tracking process, the face is cancelled.

10. An intelligent device equipped with the face recognition following method according to any one of claims 1 to 9.