WO2022133993A1 - 基于视频数据进行人脸注册的方法、装置和电子白板 - Google Patents

基于视频数据进行人脸注册的方法、装置和电子白板 Download PDF

Info

Publication number
WO2022133993A1
WO2022133993A1 PCT/CN2020/139354 CN2020139354W WO2022133993A1 WO 2022133993 A1 WO2022133993 A1 WO 2022133993A1 CN 2020139354 W CN2020139354 W CN 2020139354W WO 2022133993 A1 WO2022133993 A1 WO 2022133993A1
Authority
WO
WIPO (PCT)
Prior art keywords
face
frames
image frame
image
face detection
Prior art date
Application number
PCT/CN2020/139354
Other languages
English (en)
French (fr)
Inventor
许景涛
Original Assignee
京东方科技集团股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 京东方科技集团股份有限公司 filed Critical 京东方科技集团股份有限公司
Priority to PCT/CN2020/139354 priority Critical patent/WO2022133993A1/zh
Priority to CN202080003669.XA priority patent/CN115053268A/zh
Priority to US17/595,605 priority patent/US11908235B2/en
Publication of WO2022133993A1 publication Critical patent/WO2022133993A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/161—Detection; Localisation; Normalisation
    • G06V40/167—Detection; Localisation; Normalisation using comparisons between temporally consecutive images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/172—Classification, e.g. identification
    • G06V40/173—Classification, e.g. identification face re-identification, e.g. recognising unknown faces across different face tracks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/761—Proximity, similarity or dissimilarity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/48—Matching video sequences
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168—Feature extraction; Face representation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/40—Spoof detection, e.g. liveness detection
    • G06V40/45—Detection of the body part being alive
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/50—Maintenance of biometric data or enrolment thereof

Definitions

  • the embodiments of the present disclosure relate to the field of face recognition, and in particular, to a method, a device, and an electronic whiteboard for face registration based on video data.
  • the electronic whiteboard can receive the content written on its whiteboard and transmit the received content to the computer, so as to conveniently record and store the content on the whiteboard.
  • the function of locking the electronic whiteboard will not be provided. Therefore, anyone can modify the content on the electronic whiteboard, which leads to the problem of poor confidentiality during the use of the electronic whiteboard.
  • Embodiments of the present disclosure provide a method and device for face registration based on video data, and an electronic whiteboard.
  • a method for face registration based on video data including: receiving video data; obtaining a first image frame sequence from the video data, the first image frame sequence Each image frame in the image frame includes a face detection frame containing complete face features; determine whether the image frame reaches a preset definition according to the relative position between the face detection frames in each image frame; When the image frame reaches a preset definition, extracting multiple groups of face features based on the image information of the multiple face detection frames, and determining whether the face represents the same object according to the multiple groups of face features; and If it is determined that the human face represents the same object, the object is registered according to the first sequence of image frames.
  • obtaining the first sequence of image frames from the video data includes: obtaining a plurality of image frames from the video data in the order in which the videos are captured; determining whether the image frames contain a person based on a face detection model a face; and if it is determined that the image frame contains a human face, determining a face detection frame containing the human face in each of the plurality of image frames.
  • acquiring the first sequence of image frames from the video data further includes: determining whether the acquired image frames contain complete facial features; in the case that the image frames contain complete facial features, The image frame is stored as one frame in the first image frame sequence; when the stored first image frame sequence includes a predetermined number of frames, the acquisition of the image frame is ended.
  • determining whether the acquired image frame contains a complete face feature includes: determining whether the face is a frontal face based on a face pose detection model; determining whether the face contained in the image frame is a frontal face In this case, determine whether the face is occluded based on the face occlusion detection model; in the case of determining that the face contained in the image frame is not occluded, determine that the image frame contains a complete face feature; and otherwise , it is determined that the image frame does not contain complete facial features.
  • determining whether the image frame reaches a preset definition according to the relative positions between the face detection frames in each image frame includes: determining whether the image frame in the first image frame sequence The first ratio of the area of the intersection area of the face detection frames relative to the area of the union area of the face detection frames in the two image frames; and when the determined first ratio is greater than the first threshold , and determine that the image frame reaches the preset definition.
  • determining whether the image frame reaches a preset definition according to the relative positions between the face detection frames in each image frame includes: determining whether the image frame in the first image frame sequence The first ratio of the area of the intersection area of the face detection frames relative to the area of the union area of the face detection frames in the two image frames; determining the number of the first ratios greater than the first threshold relative to all determining a second ratio of the total number of the first ratios; and determining that the image frame reaches a preset definition when the second ratio is greater than or equal to a second threshold.
  • determining whether the human face represents the same object according to the multiple sets of human face features includes: determining a similarity between the human face features in any two adjacent image frames in the first sequence of image frames and when the determined degrees of similarity are all greater than the third threshold, it is determined that the human face represents the same object.
  • the face feature includes a face feature vector
  • determining the similarity between the face features in any two adjacent image frames in the first sequence of image frames includes: determining the distance between face feature vectors in two adjacent image frames in the first image frame sequence.
  • registering the object according to the first sequence of image frames includes registering the object with a specified image frame in the first sequence of image frames as registration data.
  • the method further includes: storing registration data obtained by registering the object according to the first image frame sequence as a face database; to identify.
  • recognizing a face in the received video data based on the face database comprises: obtaining a second sequence of image frames from the received video data, each of the second sequence of image frames Each of the image frames includes a face detection frame containing complete face features; according to the relative positions between the face detection frames in each image frame, it is determined whether the image frame includes a living human face; when it is determined that the image frame includes a living body In the case of a human face, extracting a face feature based on the face detection frame; and determining whether the face feature matches the registration data in the face database to identify the face.
  • determining whether the image frame includes a live human face according to the relative positions between the face detection frames in each image frame includes: determining that a coincidence condition is met in the face detection frames in each image frame face detection frame; determine the third ratio of the number of face detection frames that meet the coincidence condition relative to the number of all face detection frames in the face detection frame; and the third ratio is greater than or equal to
  • the fourth threshold it is determined that the human face is a non-living human face; in the case that the third ratio is smaller than the fourth threshold, it is determined that the human face is a living human face.
  • determining the face detection frame that meets the coincidence condition in the face detection frame in each image frame includes: determining the area of the intersection area between any two face detection frames in the face detection frame With respect to the fourth ratio of the area of each face detection frame in the any two face detection frames; when the determined fourth ratio is greater than the fifth threshold, it is determined that the any two face detection frames are coincident Conditional face detection frames; and when the determined fourth ratio is smaller than the fifth threshold, determine that the any two face detection frames are face detection frames that do not meet the coincidence condition.
  • determining whether the image frame includes a live human face according to relative positions between the face detection frames in each image frame further includes: one of the determined fourth ratios is greater than the determined fourth ratio. In the case where the fifth threshold is used and another fourth ratio is less than or equal to the fifth threshold, it is determined that the human face is a non-living human face.
  • an apparatus for performing face registration based on video data comprising: a memory configured to store instructions; and a processor configured to execute the instructions to perform implementations according to the present disclosure The method provided by the first aspect of the example.
  • an electronic whiteboard including the device provided according to the second aspect of the embodiments of the present disclosure.
  • the face registration can be realized without the user performing complex interaction in the registration process, thereby simplifying the registration operation steps, shortening the registration time, and improving the user experience.
  • FIG. 1 shows a flowchart of a method for face registration based on video data according to an embodiment of the present disclosure
  • FIG. 2 shows a process of acquiring a first sequence of image frames from video data according to an embodiment of the present disclosure
  • 3A and 3B respectively illustrate an example of determining whether an image frame reaches a preset definition based on relative positions between face detection frames according to an embodiment of the present disclosure
  • FIG. 4 shows an example of calculating the intersection of face detection frames based on the coordinates and dimensions of the face detection frames according to an embodiment of the present disclosure
  • FIG. 5 shows a flowchart of a method for recognizing and unlocking a face in received video data based on a face database according to another embodiment of the present disclosure
  • FIG. 6 shows a process of determining a face detection frame that meets a coincidence condition in a plurality of face detection frames according to an embodiment of the present disclosure
  • FIG. 7 shows a block diagram of an apparatus for face registration based on video data according to another embodiment of the present disclosure.
  • FIG. 8 shows a block diagram of an electronic whiteboard according to another embodiment of the present disclosure.
  • FIG. 1 shows a flowchart of a method 100 for face registration based on video data according to an embodiment of the present disclosure. As shown in FIG. 1 , the method 100 for face registration based on video data may include the following steps.
  • step S110 video data is received.
  • step S120 a first image frame sequence is obtained from the video data, and each image frame in the first image frame sequence includes a face detection frame including complete face features.
  • step S130 it is determined whether the image frame reaches a preset definition according to the relative positions between the face detection frames in each image frame.
  • step S140 when it is determined that the image frame reaches the preset definition, multiple sets of face features are extracted based on the image information of multiple face detection frames, and whether the face represents the same object is determined according to the multiple sets of face features.
  • step S150 if it is determined that the human face represents the same object, the object is registered according to the first image frame sequence.
  • video data of the object may be captured by a video capture device such as a camera.
  • the video data of the object can also be captured by a camera with a timed photographing function.
  • a video capture device or an image capture device that can obtain continuous image frames.
  • the format of the video data is not limited.
  • step S120 after the video data is received, a first sequence of image frames is obtained from the video data.
  • a face detection frame containing complete face features is included in each image frame in the first sequence of image frames.
  • Image frames that do not include a face detection frame containing complete face features in them cannot be used in the face registration process.
  • the objects may be screened according to preset rules. Ensure that only one object is registered by filtering on registered objects.
  • a plurality of image frames are obtained from the video data in the order in which the videos are captured, and based on a face detection model, it is determined whether the image frames contain a human face, and if it is determined that the image frames contain a human face, among the plurality of image frames Determine the face detection frame in each image frame of .
  • the embodiments of the present disclosure do not limit the adopted face detection model, any face detection model may be adopted, or a special detection model may be established through model training.
  • the parameters of the face detection frame may be in the form of a quaternion array, respectively recording the coordinates of the reference point and the lengths of the two sides of the face detection frame, so as to determine the position and size of the face detection frame (or face).
  • the process of screening the registered objects may include: determining a face detection frame including the face of each object in the image frame, and comparing the area of the area enclosed by each face detection frame, and selecting the enclosed area therefrom The face detection frame with the largest area, and the face contained in the face detection frame is used as the registration object.
  • a video capture window may also be provided through a Graphical User Interface (GUI) when the video capture device captures video, so as to prompt the object to place his face in the video capture window. Complete video capture.
  • GUI Graphical User Interface
  • step S130 the action behavior of the object in the plurality of image frames sequentially arranged in the first image frame sequence is analyzed by analyzing the relative positions between the face detection frames. For example, you can analyze whether the object is moving, the direction of the movement, and the magnitude of the movement. If the movement of the object is too large, it may cause the image frame captured by the video capture device to be blurred. Blurred image frames cannot be used for authentication during the registration process, nor can they be stored as final registration data for objects. Therefore, in the embodiment of the present disclosure, by analyzing the relative positions between the face detection frames to determine whether the motion range of the face is within a predetermined range, it can be determined whether the captured image frame reaches a preset definition.
  • step S140 if it can be determined that the motion range of the human face is within a predetermined range, that is, the image frame reaches a preset definition, it can be further determined based on the image frame whether the human face in each image frame belongs to the same object.
  • a face feature extraction model may be used to extract multiple sets of face features, and the extracted face features are feature vectors with a certain dimension.
  • step S150 in the case that it is ensured that a clear image frame containing complete face features has been used for registration authentication, and the faces in each image frame belong to the same object, the first image frame can be A specified image frame in the sequence is stored as the object's registration data.
  • the registration authentication process can be completed only by analyzing the received video data without requiring the registration object to cooperate with blinking, mouth opening and other interactive methods, which greatly simplifies the registration authentication process.
  • FIG. 2 illustrates a process of acquiring a first sequence of image frames from video data according to an embodiment of the present disclosure.
  • step S201 image frames are sequentially acquired from multiple image frames, the multiple image frames are consecutive image frames acquired from video data according to the capturing sequence of the video, and the extracted image frames can be Sequences are temporarily stored in the cache.
  • step S203 starting from the first frame of the plurality of image frames, the i-th image frame is sequentially acquired. Next, it is determined whether the acquired image frame contains complete facial features. This is because the model that processes the face has certain requirements on the quality of the input data. If the face in the image frame is occluded, or the face deviates greatly from the frontal posture, it is not conducive to the processing of the data by the model.
  • step S204 it is determined whether the face is a frontal face based on the face gesture detection model.
  • face key points can be obtained by training face key points using Deep Alignment Network (DAN), Tweaked Convolutional Neural Network (TCNN), etc. And the obtained face key points are input into the face pose detection model, so as to estimate the pose of the face in the image frame according to the face key points.
  • the face pose detection model can calculate the pitch angle, yaw angle and roll angle of the face separately, and determine whether the face is a straight face based on the pitch angle, yaw angle and roll angle, or whether the deflection range of the face is allowed. within the range.
  • step S205 if it is determined that the face is a frontal face, it is determined whether the face is occluded based on a face occlusion detection model. For example, it is possible to determine whether a face is occluded by using seetaface's face occlusion model. Alternatively, lightweight networks such as shuffleNet and mobileNet can be used to classify and train frontal and occluded faces to obtain a face occlusion model to determine whether a face is occluded.
  • a face occlusion detection model For example, it is possible to determine whether a face is occluded by using seetaface's face occlusion model.
  • lightweight networks such as shuffleNet and mobileNet can be used to classify and train frontal and occluded faces to obtain a face occlusion model to determine whether a face is occluded.
  • step S206 that is, in the case where it has been determined that the extracted image frame contains a frontal face and is not occluded, it is determined that the extracted image frame contains complete face features, and the extracted image frame , that is, the ith image frame is stored as a frame in the first image frame sequence S1.
  • step S207 it is determined whether a predetermined number of image frames are included in the stored first image frame sequence S1.
  • the predetermined number of image frames may be determined according to the computing capability of the computing device that performs the registration. For example, if the computing power of the computing device is relatively strong, the predetermined number of frames may be appropriately increased, for example, the predetermined number of frames may be determined to be 30 or 50 frames or more. If the computing power of the computing device is weak, the predetermined number of frames may be determined to be 20 frames or less. The predetermined number of frames may be determined by weighing the authentication accuracy requirement in the registration process, the computing power of the device, and the registration authentication time requirement.
  • the first image frame sequence obtained by the method according to the embodiment of the present disclosure includes a plurality of image frames each including complete face features, which can be used for the analysis of the action behavior of the face and the analysis of the face features during the registration process. identify.
  • determining whether an image frame reaches a preset definition according to a relative position between the face detection frames in each image frame includes: determining a face in two image frames in the first image frame sequence a first ratio of the area of the intersection area of the detection frames to the area of the union area of the face detection frames in the two image frames, and if the determined first ratios are all greater than the first threshold, determining the image The frame reaches the preset definition.
  • determining whether an image frame reaches a preset definition according to a relative position between the face detection frames in each image frame includes: determining whether an image frame of two image frames in the first sequence of image frames has a predetermined definition. A first ratio of the area of the intersection area of the face detection frames to the area of the union area of the face detection frames in the two image frames, determining the number of the first ratios greater than the first threshold relative to the total of the first ratios and if the second ratio is greater than or equal to the second threshold, it is determined that the image frame reaches the preset definition.
  • the two image frames in the first sequence of image frames used to perform the calculation may be adjacent image frames or may be spaced image frames.
  • the first image frame sequence S1 include image frames F 1 , F 2 , F 3 , F 4 , F 5 , F 6 , etc. image frames.
  • the first ratio may be calculated between F1 and F2 , between F2 and F3, and between F3 and F4 , respectively Calculate the first ratio between, . . . and so on.
  • the calculation may be performed one image frame spaced apart, eg, the first ratio is calculated between F1 and F3 , and between F3 and F5 , respectively The first ratio, ... and so on.
  • the calculation may be performed with two or more image frames spaced apart, for example calculating the first ratio between F 1 and F 4 , . . . and so on analogy.
  • 3A and 3B respectively illustrate an example of determining whether an image frame reaches a preset definition based on relative positions between face detection frames according to an embodiment of the present disclosure.
  • FIGS. 3A and 3B only the case where the first ratio is calculated for adjacent image frames is described as an example.
  • the first image frame sequence includes multiple image frames, and the ratio of the area of the intersection area and the area of the union area of the face detection frames in the two image frames is calculated to analyze the action behavior of the object.
  • the ratio of the area of the intersection area and the area of the union area between two adjacent face detection frames can be calculated as F 12 /(F 1 +F 2 -F 12 ), where F 1 represents the first The face detection frame in the first image frame, F 2 represents the face detection frame in the second image frame, and at the same time, F 1 and F 2 represent the area of the face detection frame F 1 and F 2 , which is represented by F 12 .
  • the first threshold value may be set according to the reliability requirements and image clarity requirements of the registration. If the first threshold is set to a larger value, the quality of the image can be improved, that is, to ensure that the image is clearer, but it may lead to the failure to continue the authentication for multiple registrations. Conversely, if the first threshold is set to a smaller value, the registration authentication can be performed more smoothly, but it may introduce more unclear images, thereby affecting the reliability of the registration authentication. According to an embodiment, the quality of the image may be guaranteed by adjusting the first threshold.
  • the process of calculating the ratio of the area of the intersection area and the area of the union area of the face detection frames in adjacent image frames is the same as that shown in FIG. 3A, and F12 can be calculated with reference to FIG. 3A. /(F 1 +F 2 -F 12 ).
  • the number N 1 of the first ratios greater than the first threshold is counted, and then the second ratio N 1 of the number N 1 of the first ratios greater than the first threshold relative to the total number N of the first ratios is calculated /N. If N 1 /N is greater than or equal to the second threshold, it is determined that the image frame reaches the preset definition.
  • the image frames that reach the preset definition reach a certain scale, that is, the ratio of the number N1 of the first ratio greater than the first threshold to the total number N of the first ratios reaches a certain requirement, that is, the When the second ratio N 1 /N of the number N 1 of the first ratio greater than the first threshold relative to the total number N of the first ratios is greater than or equal to the second threshold, it is considered that the image frame reaches the preset definition.
  • the quality of the image may be guaranteed by adjusting the first threshold and the second threshold in coordination. By introducing two adjustment parameters, it is more flexible and accurate to judge whether the image frame reaches the preset definition.
  • FIG. 4 shows an example of calculating the intersection of face detection frames based on the coordinates and sizes of the face detection frames according to an embodiment of the present disclosure.
  • the coordinate system above Figure 4 is a coordinate system established with the upper left corner of the image frame as the coordinate origin, the positive direction of the X axis is the direction extending along one side of the image frame, and the positive direction of the Y axis is the direction along the image frame. The direction in which the other side of the frame extends.
  • the position and size of the face detection frame in the first image frame can be represented by a parameter set [x 1 , y 1 , w 1 , h 1 ].
  • x 1 and y 1 represent the coordinates of the upper left corner of the face detection frame
  • w 1 represents the length of the face detection frame along the X-axis direction
  • h 1 represents the length of the face detection frame along the Y-axis direction. Shown below the coordinate system is the process of intersecting the face detection frame in the first image frame and the face detection frame in the second image frame. As shown in FIG.
  • determining whether a face represents the same object according to multiple sets of face features includes determining the similarity between the face features in any two adjacent image frames in the first image frame sequence, and determining the similarity between the determined similarity If the degrees are all greater than the third threshold, it is determined that the human face represents the same object; otherwise, it is determined that the human face represents different objects.
  • the facial features can be acquired by invoking the facial feature extraction model. Different facial feature extraction models output feature vectors of different dimensions. For the feature vector, the similarity between the facial features in any two adjacent image frames in the first image frame sequence can be determined by calculating the distance between the feature vectors.
  • the setting of the third threshold may be determined according to the database used by the adopted facial feature extraction model. Different facial feature extraction models will give the recognition accuracy and corresponding threshold settings. If it is determined through analysis and identification that the faces in each image frame in the first image frame sequence belong to the same object, the specified image frame in the first image frame sequence can be used as the registration data registration object.
  • the registration data before the registration data is saved, the registration data may be compared with the registration data previously saved in the face database, and if the face has already been registered, the storage may not be overwritten.
  • the definition of the image frame can be determined by using the video for registration and analyzing the relative positions between the face detection frames in the multiple image frames, without the need for the user to cooperate with operations such as blinking, opening the mouth, etc. , thereby simplifying the process of registration authentication and ensuring the reliability of registration data.
  • FIG. 5 shows a flowchart of a method 500 for recognizing and unlocking a face in received video data based on a face database according to another embodiment of the present disclosure. As shown in Figure 5, method 500 includes the following steps:
  • step S510 input video frame data is received.
  • step S520 a second image frame sequence is obtained from the received video data, and each image frame in the second image frame sequence includes a face detection frame including complete face features.
  • step S530 it is determined whether the image frame includes a living human face according to the relative positions between the face detection frames in each image frame.
  • step S540 if it is determined that the image frame includes a living human face, face features are extracted based on the face detection frame.
  • step S550 it is determined whether the face features match the registration data in the face database, so as to identify the face.
  • step S560 unlocking is identified.
  • steps S510 , S520 , S540 and S550 can be obtained from steps S110 , S120 and S140 in the method 100 for face registration based on video data in the foregoing embodiment, which will not be repeated here.
  • determining whether an image frame includes a live human face according to the relative positions between the face detection frames in each image frame includes determining, in the face detection frame in each image frame, a face detection that meets the coincidence condition frame, determine a third ratio of the number of face detection frames that meet the coincidence condition relative to the number of all face detection frames in multiple face detection frames, and when the third ratio is greater than or equal to the fourth threshold, determine the person The face is a non-living human face, and in the case that the third ratio is smaller than the fourth threshold, it is determined that the human face is a living human face.
  • determining the face detection frame that meets the coincidence condition in the face detection frame in each image frame includes determining the area of the intersection area between any two face detection frames in the face detection frame relative to the arbitrary The fourth ratio of the area of each face detection frame in the two face detection frames, in the case where the determined fourth ratio is greater than the fifth threshold, it is determined that any two face detection frames are face detection frames that meet the coincidence condition , and when the determined fourth ratio is smaller than the fifth threshold, it is determined that any two face detection frames are face detection frames that do not meet the coincidence condition.
  • an intersection operation is performed between any two face detection frames in a plurality of face detection frames, and the area of the calculated intersection area and each face detection frame in the two face detection frames in which the intersection operation is performed are calculated.
  • the ratio of the areas of , and the degree of coincidence between the two face detection frames can be determined by the obtained ratio.
  • a fifth threshold is set to measure the degree of coincidence between the two face detection frames. If the fifth threshold is set higher, the two face detection frames can only be determined to be coincident when the degree of coincidence of the two face detection frames is high.
  • the coincidence of the face detection frames means that the object is in the time period between the two face detection frames.
  • the fifth threshold is set higher, the proportion of the overlapping face detection frames in all the face detection frames will be reduced, and the possibility that a non-living body will be recognized as a living body will be increased.
  • the fifth threshold is set lower, more face detection frames will be determined to be coincident, which will increase the possibility of a living body being identified as a non-living body.
  • the fifth threshold may be set according to the occasion of registering the authentication application.
  • the fifth threshold may be set higher, because in these occasions, it is basically guaranteed that the object is a living body, so it is less likely that the living body is identified as a non-living body The possibility of a live object can be fully guaranteed to be correctly identified, thereby improving the user experience.
  • Whether the face is a live face can be determined by analyzing the action behavior of the object. That is, in the embodiment of the present disclosure, whether the object is a living body can be determined only by analyzing the relative positions between the face detection frames in the multiple image frames, thereby effectively preventing the unlocking based on the non-living video. Operations, such as unlocking operations using a photo of an object can be avoided, improving locking security.
  • FIG. 6 shows a process of determining a face detection frame that meets a coincidence condition in a plurality of face detection frames according to an embodiment of the present disclosure.
  • FIG. 6 shows a process of determining a face detection frame that meets a coincidence condition in a plurality of face detection frames according to an embodiment of the present disclosure.
  • FIG. 6 shows a process of determining a face detection frame that meets a coincidence condition in a plurality of face detection frames according to an embodiment of the present disclosure.
  • the face is determined to be a non-living face if one of the second ratios is less than or equal to the second threshold and the other second ratio is greater than the second threshold.
  • the face detection frames eg, F 1 , F 2 , F 7
  • the sizes of the face detection frames may be different from each other, but the sizes are not much different from each other.
  • the size of a face detection frame is significantly different from the size of other face detection frames, it means that the face contained in the face detection frame may have a large range of motion, or the face contained in the face detection frame may have a large range of motion.
  • the face may not belong to the same person as the faces contained in other face detection frames. Therefore, in this case, it can be directly determined that the object is a non-living body according to the comparison result between the fourth ratio and the fifth threshold, and it is no longer necessary to continue to determine whether other face detection frames meet the coincidence condition.
  • FIG. 7 shows a block diagram of an apparatus 700 for registration based on video data according to another embodiment of the present disclosure.
  • the apparatus 700 includes a processor 701 , a memory 702 and a camera 703 , and machine-readable instructions are stored in the memory 702 , and the processor 701 can execute these machine-readable instructions to implement the based on the embodiments of the present disclosure.
  • the camera 703 may be configured to acquire video data, and the frame rate of the camera 703 may be in the range of 15-25 frames per second.
  • Memory 702 may be in the form of non-volatile or volatile memory, eg, electrically erasable programmable read only memory (EEPROM), flash memory, and the like.
  • EEPROM electrically erasable programmable read only memory
  • Various components inside the apparatus 700 may be implemented by various devices, including but not limited to: analog circuit devices, digital circuit devices, digital signal processing (DSP) circuits, programmable processors, dedicated Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs), Programmable Logic Devices (CPLDs), etc.
  • DSP digital signal processing
  • ASICs dedicated Integrated Circuits
  • FPGAs Field Programmable Gate Arrays
  • CPLDs Programmable Logic Devices
  • FIG. 8 shows a block diagram of an electronic whiteboard 800 according to another embodiment of the present disclosure.
  • an electronic whiteboard 800 according to an embodiment of the present disclosure includes a display whiteboard 801 and an apparatus 802 for performing face registration based on video data according to an embodiment of the present disclosure.
  • a device for registration based on video data is installed, and face registration is directly performed by intercepting a video stream without human interaction, and registration by directly acquiring video frames is more convenient.
  • the electronic whiteboard according to the embodiment of the present disclosure does not need to be switched on and off manually, and can be directly unlocked and used through face information within a certain distance, and the confidentiality is good. Moreover, only the fixed face registered by appointment can be unlocked, which effectively protects the information security of the appointment user during the use of the electronic whiteboard.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • General Physics & Mathematics (AREA)
  • Human Computer Interaction (AREA)
  • Health & Medical Sciences (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • Medical Informatics (AREA)
  • Image Analysis (AREA)

Abstract

本公开实施例提供了一种基于视频数据进行人脸注册的方法，包括：接收视频数据；从视频数据中获取第一图像帧序列，第一图像帧序列中的每个图像帧各自包括包含完整人脸特征的人脸检测框；根据每个图像帧中的人脸检测框之间的相对位置确定图像帧是否达到预设清晰度；在确定图像帧达到预设清晰度的情况下，基于多个人脸检测框的图像信息提取多组人脸特征，并根据多组人脸特征确定人脸是否表示同一对象；以及在确定人脸表示同一对象的情况下，根据第一图像帧序列注册对象。本公开实施例还提供了一种基于视频数据进行人脸注册的装置和电子白板。

Description

基于视频数据进行人脸注册的方法、装置和电子白板 技术领域
本公开实施例涉及人脸识别领域，尤其涉及一种基于视频数据进行人脸注册的方法、装置和电子白板。
背景技术
随着无纸化会议和无纸化办公的逐渐普及，电子白板的应用也越来越广泛。电子白板可以接收书写在其白板板面上的内容并将接收到的内容传输至计算机，从而方便地对白板板面上的内容进行记录和存储。在使用电子白板时，为了能够在任何距离处都方便地对电子白板进行操作，不会设置对电子白板进行锁定的功能。因此，任何人都可以对电子白板上的内容进行修改，这导致电子白板在使用过程中存在保密性差的问题。
发明内容
本公开实施例提供了一种基于视频数据进行人脸注册的方法、装置以及一种电子白板。
根据本公开实施例的第一方面，提供了一种基于视频数据进行人脸注册的方法，包括：接收视频数据；从所述视频数据中获取第一图像帧序列，所述第一图像帧序列中的每个图像帧各自包括包含完整人脸特征的人脸检测框；根据每个图像帧中的人脸检测框之间的相对位置确定所述图像帧是否达到预设清晰度；在确定所述图像帧达到预设清晰度的情况下，基于所述多个人脸检测框的图像信息提取多组人脸特征，并根据所述多组人脸特征确定所述人脸是否表示同一对象；以及在确定所述人脸表示同一对象的情况下，根据所述第一图像帧序列注册所述对象。
在一些实施例中，从所述视频数据中获取第一图像帧序列包括：按照视频的捕获顺序从所述视频数据中获取多个图像帧；基于人脸检测模型确定所述图像帧是否包含人脸；以及在确定所述图像帧包含人脸的情况下，在所述多个图像帧中的每个图像帧中确定包含所述人脸的人脸检测框。
在一些实施例中，从所述视频数据中获取第一图像帧序列还包括：确定所获取的图像帧是否包含完整人脸特征；在所述图像帧包含完整人脸特征的情况下，将所述图 像帧存储为第一图像帧序列中的一帧；在已存储的第一图像帧序列包括预定帧数的情况下，结束获取图像帧。
在一些实施例中，确定所获取的图像帧是否包含完整人脸特征包括：基于人脸姿态检测模型确定所述人脸是否是正脸；在确定所述图像帧中所包含的人脸是正脸的情况下，基于人脸遮挡检测模型确定所述人脸是否被遮挡；在确定所述图像帧中所包含的人脸未被遮挡的情况下，确定所述图像帧包含完整人脸特征；以及否则，确定所述图像帧不包含完整人脸特征。
在一些实施例中，根据每个图像帧中的人脸检测框之间的相对位置确定所述图像帧是否达到预设清晰度包括：确定所述第一图像帧序列中的两个图像帧中的人脸检测框的交集区域的面积相对于该两个图像帧中的人脸检测框的并集区域的面积的第一比率；以及在所确定的第一比率均大于第一阈值的情况下，确定所述图像帧达到预设清晰度。
在一些实施例中，根据每个图像帧中的人脸检测框之间的相对位置确定所述图像帧是否达到预设清晰度包括：确定所述第一图像帧序列中的两个图像帧中的人脸检测框的交集区域的面积相对于该两个图像帧中的人脸检测框的并集区域的面积的第一比率；确定大于第一阈值的所述第一比率的数量相对于所述第一比率总的数量的第二比率；以及在所述第二比率大于或等于第二阈值的情况下，确定所述图像帧达到预设清晰度。
在一些实施例中，根据所述多组人脸特征确定所述人脸是否表示同一对象包括：确定所述第一图像帧序列中任意相邻两个图像帧中的人脸特征之间的相似度；以及在所确定的相似度均大于第三阈值的情况下，确定所述人脸表示同一对象。
在一些实施例中，所述人脸特征包括人脸特征向量，并且其中，确定所述第一图像帧序列中任意相邻两个图像帧中的人脸特征之间的相似度包括：确定所述第一图像帧序列中相邻两个图像帧中的人脸特征向量之间的距离。
在一些实施例中，根据所述第一图像帧序列注册所述对象包括：以所述第一图像帧序列中的指定图像帧作为注册数据注册所述对象。
在一些实施例中，方法还包括：将根据所述第一图像帧序列注册所述对象而得到的注册数据存储为人脸库；以及基于所述人脸库对所接收的视频数据中的人脸进行识别。
在一些实施例中，基于所述人脸库对所接收的视频数据中的人脸进行识别包括：从接收的视频数据中获取第二图像帧序列，所述第二图像帧序列中的每个图像帧各自包括包含完整人脸特征的人脸检测框；根据每个图像帧中的人脸检测框之间的相对位置确定所述图像帧是否包括活体人脸；在确定所述图像帧包括活体人脸的情况下，基于所述人脸检测框提取人脸特征；以及确定所述人脸特征是否与所述人脸库中的注册数据相匹配，以识别所述人脸。
在一些实施例中，根据每个图像帧中的人脸检测框之间的相对位置确定所述图像帧是否包括活体人脸包括：在每个图像帧中的人脸检测框中确定符合重合条件的人脸检测框；确定所述符合重合条件的人脸检测框的数量相对于所述人脸检测框中全部人脸检测框的数量的第三比率；以及在所述第三比率大于或等于第四阈值的情况下，确定所述人脸是非活体人脸；在所述第三比率小于所述第四阈值的情况下，确定所述人脸是活体人脸。
在一些实施例中，在每个图像帧中的人脸检测框中确定符合重合条件的人脸检测框包括：确定所述人脸检测框中任意两个人脸检测框之间的交集区域的面积相对于该任意两个人脸检测框中每个人脸检测框的面积的第四比率；在所确定的第四比率均大于第五阈值的情况下，确定所述任意两个人脸检测框是符合重合条件的人脸检测框；以及在所确定的第四比率均小于第五阈值的情况下，确定所述任意两个人脸检测框是不符合重合条件的人脸检测框。
在一些实施例中，根据每个图像帧中的人脸检测框之间的相对位置确定所述图像帧是否包括活体人脸还包括：在所确定的第四比率中的一个第四比率大于所述第五阈值且另一个第四比率小于或等于所述第五阈值的情况下，确定所述人脸是非活体人脸。
根据本公开实施例的第二方面，提供了一种基于视频数据进行人脸注册的装置，包括：存储器，配置为存储指令；以及处理器，配置为执行所述指令，以执行根据本公开实施例的第一方面所提供的方法。
根据本公开实施例的第三方面，提供了一种电子白板，包括根据本公开实施例的第二方面提供的装置。
根据本公开实施例的基于视频数据进行人脸注册的方法，无需用户在注册过程中进行复杂的交互就能够实现人脸的注册，从而简化注册操作的步骤，缩短注册时间，提升了用户体验。
附图说明
为了更清楚地说明本公开实施例的技术方案，下面将对本公开实施例描述中所需要使用的附图作简单地介绍。显而易见地，下面描述中的附图仅仅是本公开的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图，在附图中：
图1示出了根据本公开实施例的基于视频数据进行人脸注册的方法的流程图；
图2示出了根据本公开实施例的从视频数据中获取第一图像帧序列的过程；
图3A和图3B分别示出了根据本公开实施例的基于人脸检测框之间的相对位置确定图像帧是否达到预设清晰度的示例；
图4示出了根据本公开实施例的基于人脸检测框的坐标和尺寸计算人脸检测框的交集的示例；
图5示出了根据本公开另一实施例的基于人脸库对所接收的视频数据中的人脸进行识别解锁的方法的流程图；
图6示出了根据本公开实施例的在多个人脸检测框中确定符合重合条件的人脸检测框的过程；
图7示出了根据本公开另一实施例的基于视频数据进行人脸注册的装置的框图；以及
图8示出了根据本公开另一实施例的电子白板的框图。
具体实施方式
为使本公开实施例的目的、技术方案和优点更加清楚，下面将结合本公开实施例中的附图，对本公开实施例中的技术方案进行清楚、完整的描述。显然，所描述的实施例是本公开的一部分实施例，而不是全部。基于所描述的本公开实施例，本领域普通技术人员在无需创造性劳动的前提下获得的所有其他实施例都属于本公开保护的范围。应注意，贯穿附图，相同的元素由相同或相近的附图标记来表示。在以下描述中，一些具体实施例仅用于描述目的，而不应该理解为对本公开有任何限制，而只是本公开实施例的示例。在可能导致对本公开的理解造成混淆时，将省略常规结构或构造。应注意，图中各部件的形状和尺寸不反映真实大小和比例，而仅示意本公开实施例的内容。
除非另外定义，本公开实施例使用的技术术语或科学术语应当是本领域技术人员所理解的通常意义。本公开实施例中使用的“第一”、“第二”以及类似词语并不表示任何顺序、数量或重要性，而只是用于区分不同的组成部分。
图1示出了根据本公开实施例的基于视频数据进行人脸注册的方法100的流程图。如图1所示，基于视频数据进行人脸注册的方法100可以包括以下步骤。
在步骤S110中，接收视频数据。
在步骤S120中，从视频数据中获取第一图像帧序列，第一图像帧序列中的每个图像帧各自包括包含完整人脸特征的人脸检测框。
在步骤S130中，根据每个图像帧中的人脸检测框之间的相对位置确定图像帧是否达到预设清晰度。
在步骤S140中，在确定图像帧达到预设清晰度的情况下，基于多个人脸检测框的图像信息提取多组人脸特征，并根据多组人脸特征确定人脸是否表示同一对象。
在步骤S150中，在确定人脸表示同一对象的情况下，根据第一图像帧序列注册对象。
根据实施例，在步骤S110中，可以通过诸如摄像头之类的视频采集设备来捕获对象的视频数据。在其他的实施例中，也可以通过具有定时拍照功能的相机来捕获对象的视频数据。只要是能够获得连续图像帧的视频采集设备或图像采集设备即可。此外，在本公开的实施例中，对视频数据的格式不做限制。
根据实施例，在步骤S120中，在接收到视频数据之后，从视频数据中获取第一图像帧序列。在第一图像帧序列中的每个图像帧中包括包含完整人脸特征的人脸检测框。未在其中包括包含完整人脸特征的人脸检测框的图像帧不能用于人脸注册过程。
根据实施例，如果视频采集设备在一个图像帧中捕获到多个对象，则可以根据预设的规则对对象进行筛选。通过对注册对象进行筛选，确保仅针对一个对象进行注册。根据实施例，按照视频的捕获顺序从视频数据中获取多个图像帧，并基于人脸检测模型确定图像帧是否包含人脸，在确定图像帧包含人脸的情况下，在多个图像帧中的每个图像帧中确定人脸检测框。本公开实施例对所采用的人脸检测模型不做限制，可以采用任何人脸检测模型，也可以通过模型训练建立专门的检测模型。人脸检测框的参数可以具有四元数组的形式，分别记录人脸检测框的参考点的坐标和两个边长，以用于确定人脸检测框(或人脸)的位置和尺寸。根据实施例，对注册对象进行筛选的过程可以包括：确 定包括图像帧中的每个对象人脸的人脸检测框，并分别比较各人脸检测框所包围区域的面积，从中选择所包围区域的面积最大的人脸检测框，并将该人脸检测框所包含的人脸作为注册对象。在本公开的其他实施例中，也可以在视频采集设备捕获视频时，通过图形用户界面(Graphical User Interface，GUI)提供视频捕获窗口，以提示对象将自己的脸置于该视频捕获窗口中来完成视频采集。
根据实施例，在步骤S130中，通过分析人脸检测框之间的相对位置来分析对象在第一图像帧序列中顺序排列的多个图像帧中所表现的动作行为。例如，可以分析对象是否在运动、运动的方向以及运动的幅度等。如果对象的运动幅度过大，则有可能导致视频采集设备所捕获的图像帧模糊。模糊的图像帧不能用于注册过程中的认证，也不能作为对象的最终注册数据进行存储。因此，在本公开的实施例中，通过分析人脸检测框之间的相对位置来确定人脸的运动幅度是否在预定范围内，可以确定所捕获的图像帧是否达到预设清晰度。
根据实施例，在步骤S140中，如果能够确定人脸的运动幅度在预定范围内，即图像帧达到预设清晰度，则可以基于该图像帧进一步确定各图像帧中的人脸是否属于同一个对象。根据实施例，可以利用人脸特征提取模型来提取多组人脸特征，所提取得到的人脸特征是具有一定维度的特征向量。
根据实施例，在步骤S150中，在确保已经使用了包含完整人脸特征的清晰图像帧进行注册认证，且各图像帧中的人脸都属于同一个对象的情况下，可以将第一图像帧序列中的一个指定图像帧作为对象的注册数据进行存储。
根据本公开的实施例，可以仅通过对接收的视频数据进行分析，而不需要注册对象配合眨眼、张嘴等交互方式就能够完成注册认证过程，极大地简化了注册认证的过程。
图2示出了根据本公开实施例的从视频数据中获取第一图像帧序列的过程。如图2所示，在步骤S201中，从多个图像帧中依次获取图像帧，该多个图像帧是按照视频的捕获顺序从视频数据中获取的连续图像帧，可以将提取到的图像帧序列暂时存储在缓存中。
接下来，在步骤S202中，可以设置用于提取第一图像帧序列的参数，包括设置循环变量i的初始值i＝1。
接下来，在步骤S203中，从多个图像帧中的第1帧开始，依次获取第i个图像帧。接下来，确定所获取的图像帧是否包含完整的人脸特征。这是因为对人脸进行处理的模 型对输入数据的质量是有一定要求的。如果图像帧中的人脸被遮挡，或者人脸较大地偏离正脸姿态，都不利于模型对数据的处理。
接下来，在步骤S204中，基于人脸姿态检测模型确定人脸是否是正脸。例如，可以通过使用深度对齐网络(Deep Alignment Network，DAN)、微调卷积神经网络(Tweaked Convolutional Neural Network，TCNN)等对人脸关键点进行训练来得到人脸关键点。并且将所得到的人脸关键点输入到人脸姿态检测模型中，以根据人脸关键点对图像帧中的人脸的姿态进行估计。人脸姿态检测模型可以分别计算人脸的俯仰角、偏航角和翻滚角，并基于俯仰角、偏航角和翻滚角确定人脸是否为正脸，或者人脸的偏转范围是否在允许的范围内。
接下来，在步骤S205中，在确定人脸是正脸的情况下，基于人脸遮挡检测模型确定人脸是否被遮挡。例如，可以通过使用seetaface的人脸遮挡模型来确定人脸是否被遮挡。或者，也可以使用shuffleNet、mobileNet等轻量级网络对正脸和遮挡人脸进行分类训练以得到人脸遮挡模型来确定人脸是否被遮挡。
接下来，在步骤S206中，即在已经确定在所提取的图像帧中包含正脸且未被遮挡的情况下，确定所提取的图像帧中包含完整人脸特征，并将所提取的图像帧，即第i个图像帧存储为第一图像帧序列S1中的一帧。
接下来，在步骤S207中，确定已存储的第一图像帧序列S1中是否包括预定帧数的图像帧。这里，预定帧数的图像帧可以根据执行注册的计算设备的计算能力来确定。例如，如果计算设备的计算能力较强，则可以适当增加预定帧数，例如可以将预定帧数确定为30或50帧或者更多。如果计算设备的计算能力较弱，则可以将预定帧数确定为20帧或者更少。可以权衡注册过程中的认证准确率要求、设备的计算能力和注册认证时间要求来确定预定帧数。如果确定第一图像帧序列S1中已经存储了预定帧数的图像帧，则退出继续提取图像帧的过程，得到包括预定帧数的多个图像帧的第一图像帧序列S1。如果确定第一图像帧序列S1中尚未存储预定帧数的图像帧，则在步骤S208中，将循环变量i加1，即令i＝i+1后返回到步骤S203，继续从多个图像帧中获取第i个图像帧，直到第一图像帧序列S1中存储了预定帧数的图像帧。
根据本公开实施例的方法得到的第一图像帧序列，包括各自包括完整的人脸特征的多个图像帧，可以将其用于注册过程中针对人脸的动作行为的分析和人脸特征的识别。
根据本公开的实施例，根据每个图像帧中的人脸检测框之间的相对位置确定图像 帧是否达到预设清晰度包括：确定第一图像帧序列中的两个图像帧中的人脸检测框的交集区域的面积相对于该两个图像帧中的人脸检测框的并集区域的面积的第一比率，以及在所确定的第一比率均大于第一阈值的情况下，确定图像帧达到预设清晰度。
根据本公开的另一实施例，根据每个图像帧中的人脸检测框之间的相对位置确定图像帧是否达到预设清晰度包括：确定第一图像帧序列中的两个图像帧中的人脸检测框的交集区域的面积相对于该两个图像帧中的人脸检测框的并集区域的面积的第一比率，确定大于第一阈值的第一比率的数量相对于第一比率总的数量的第二比率，以及在第二比率大于或等于第二阈值的情况下，确定图像帧达到预设清晰度。
根据实施例，用于执行计算的第一图像帧序列中的两个图像帧可以是相邻的图像帧，也可以是间隔的图像帧。例如，令第一图像帧序列S1包括图像帧F 1、F 2、F 3、F 4、F 5、F 6……等图像帧。在针对相邻的图像帧计算第一比率的实施例中，可以分别在F 1和F 2之间计算第一比率，在F 2和F 3之间计算第一比率，在F 3和F 4之间计算第一比率，……并以此类推。在针对间隔的图像帧计算第一比率的另一实施例中，可以间隔一个图像帧来执行计算，例如分别在F 1和F 3之间计算第一比率，在F 3和F 5之间计算第一比率，……并以此类推。在针对间隔的图像帧计算第一比率的又一实施例中，可以间隔两个或更多个图像帧来执行计算，例如在F 1和F 4之间计算第一比率，……并以此类推。
图3A和图3B分别示出了根据本公开实施例的基于人脸检测框之间的相对位置确定图像帧是否达到预设清晰度的示例。在图3A和图3B中，仅以针对相邻的图像帧计算第一比率的情况作为示例进行说明。
如图3A所示，第一图像帧序列包括多个图像帧，计算两个图像帧中的人脸检测框的交集区域的面积与并集区域的面积之比，以分析对象的动作行为。如图3A所示，相邻两个人脸检测框的交集区域的面积与并集区域的面积之比可以计算为F 12/(F 1+F 2-F 12)，其中，F 1表示第1个图像帧中的人脸检测框，F 2表示第2个图像帧中的人脸检测框，并同时以F 1和F 2表示人脸检测框F 1和F 2的面积，以F 12表示人脸检测框F 1和F 2的交集区域的面积。
根据实施例，第一阈值可以根据注册的可靠性要求与图像清晰度要求进行设置。如果第一阈值设置得较大，则可以提高图像的质量，即确保图像更清晰，但有可能导致多次注册认证不能继续进行。反之，如果第一阈值设置得较小，则可以使注册认证的进行 更流畅，但有可能引入比较多的不清晰图像，从而影响注册认证的可靠性。根据实施例，可以通过调整第一阈值来保证图像的质量。
如图3B所示，计算相邻图像帧中的人脸检测框的交集区域的面积与并集区域的面积之比的过程与图3A中所示的过程相同，可以参考图3A计算得到F 12/(F 1+F 2-F 12)。在图3B中，对大于第一阈值的第一比率的数量N 1进行统计，然后计算大于第一阈值的第一比率的数量N 1相对于第一比率总的数量N的第二比率N 1/N。如果N 1/N大于或等于第二阈值，则确定图像帧达到预设清晰度。
在该实施例中，即使部分图像帧的清晰度未达到预设的第一阈值，例如F 23/(F 2+F 3-F 23)小于第一阈值，也不因此而认为图像帧未达到预设清晰度。根据实施例，在达到预设清晰度的图像帧达到一定规模时，即大于第一阈值的第一比率的数量N 1在第一比率总的数量N中所占的比率达到一定要求，即在大于第一阈值的第一比率的数量N 1相对于第一比率总的数量N的第二比率N 1/N大于或等于第二阈值的情况下，就认为图像帧达到预设清晰度。根据实施例，可以通过协调调整第一阈值和第二阈值来保证图像的质量。通过引入两个调整参数，使得判断图像帧是否达到预设清晰度时更加灵活和准确。
图4示出了根据本公开实施例的基于人脸检测框的坐标和尺寸计算人脸检测框的交集的示例。如图4所示，图4上方的坐标系是以图像帧的左上角点为坐标原点建立的坐标系，X轴正方向为沿图像帧的一个边延伸的方向，Y轴正方向为沿图像帧的另一个边延伸的方向。如图4所示，可以以参数集合[x 1,y 1,w 1,h 1]来表示第1个图像帧中人脸检测框的位置和尺寸。其中，x 1和y 1表示人脸检测框左上角点的坐标，w 1表示人脸检测框沿X轴方向的长度，h 1表示人脸检测框沿Y轴方向的长度。在该坐标系下方示出的是第1个图像帧中的人脸检测框与第2个图像帧中的人脸检测框做交集的过程。如图4所示，可以确定交集区域的左上角点坐标分别为x min＝max(x 1,x 2)，y min＝max(y 1,y 2)，并且可以确定交集区域的右下角点坐标分别为x max＝min(x 1+w 1,x 2+w 2)，y max＝min(y 1+h 1,y 2+h 2)。可以根据交集区域左上角点坐标和右下角点坐标计算交集区域的面积为S 12＝(x max-x min)*(y max-y min)。
根据实施例，根据多组人脸特征确定人脸是否表示同一对象包括，确定第一图像帧序列中任意相邻两个图像帧中的人脸特征之间的相似度，以及在所确定的相似度均大于第三阈值的情况下，确定人脸表示同一对象；否则，确定人脸表示不同对象。在本公开的实施例中，可以通过调用人脸特征提取模型来获取人脸特征。不同的人脸特征提取模型输出不同维度的特征向量。对于特征向量，可以通过计算特征向量之间的距离来确定 第一图像帧序列中任意相邻两个图像帧中的人脸特征之间的相似度。根据实施例，可以采用欧式距离
Figure PCTCN2020139354-appb-000001
曼哈顿距离c＝|m i-n i|、或马氏距离
Figure PCTCN2020139354-appb-000002
Figure PCTCN2020139354-appb-000003
等计算特征向量之间的距离，其中m i和n i均表示向量。根据实施例，第三阈值的设置可以根据所采用的人脸特征提取模型所使用的数据库来确定。不同的人脸特征提取模型会给出识别准确率与相应阈值的设置。如果通过分析识别，确定第一图像帧序列中各图像帧中的人脸属于同一个对象，则可以以第一图像帧序列中的指定图像帧作为注册数据注册对象。
根据实施例，在保存注册数据之前，可以将注册数据与先前保存在人脸库中的注册数据进行相似度比较，如果该人脸已经注册，因此可以不覆盖存储。
根据本公开的实施例，通过使用视频进行注册，并通过对多个图像帧中人脸检测框之间的相对位置进行分析就可以确定图像帧的清晰度，不需要用户配合眨眼，张嘴等操作，由此简化了注册认证的过程，且保证了注册数据的可靠性。
图5示出了根据本公开另一实施例的基于人脸库对所接收的视频数据中的人脸进行识别解锁的方法500的流程图。如图5所示，方法500包括以下步骤：
在步骤S510中，接收输入视频帧数据。
在步骤S520中，从接收的视频数据中获取第二图像帧序列，第二图像帧序列中的每个图像帧各自包括包含完整人脸特征的人脸检测框。
在步骤S530中，根据每个图像帧中的人脸检测框之间的相对位置确定图像帧是否包括活体人脸。
在步骤S540中，在确定图像帧包括活体人脸的情况下，基于人脸检测框提取人脸特征。
在步骤S550中，确定人脸特征是否与人脸库中的注册数据相匹配，以识别人脸。
在步骤S560中，识别解锁。
其中，步骤S510、S520、S540和S550的操作可以参加前述实施例中的基于视频数据进行人脸注册的方法100中的步骤S110、S120和S140获得，此处不再赘述。
根据实施例，根据每个图像帧中的人脸检测框之间的相对位置确定图像帧是否包括活体人脸包括，在每个图像帧中的人脸检测框中确定符合重合条件的人脸检测框，确定符合重合条件的人脸检测框的数量相对于多个人脸检测框中全部人脸检测框的数量的第三比率，以及在第三比率大于或等于第四阈值的情况下，确定人脸是非活体人 脸，在第三比率小于第四阈值的情况下，确定人脸是活体人脸。
根据实施例，在每个图像帧中的人脸检测框中确定符合重合条件的人脸检测框包括，确定人脸检测框中任意两个人脸检测框之间的交集区域的面积相对于该任意两个人脸检测框中每个人脸检测框的面积的第四比率，在所确定的第四比率均大于第五阈值的情况下，确定任意两个人脸检测框是符合重合条件的人脸检测框，以及在所确定的第四比率均小于第五阈值的情况下，确定任意两个人脸检测框是不符合重合条件的人脸检测框。
在本实施例中，在多个人脸检测框中任意两个人脸检测框两两之间执行交集运算，并计算所得交集区域的面积与执行交集运算的两个人脸检测框中每个人脸检测框的面积之比，可以通过所得比值确定这两个人脸检测框之间的重合程度。根据实施例，设置第五阈值来衡量两个人脸检测框之间的重合程度。如果将第五阈值设置得较高，则必须在两个人脸检测框重合程度较高的情况下才能被确定为重合，人脸检测框重合表示对象在两个人脸检测框之间的时间段内大概率是没有动作行为的，即可以认为对象是静止的，进一步地，认为对象不是活体。因此如果将第五阈值设置得较高，将会降低重合的人脸检测框在全部人脸检测框之中的占比，会增加非活体被识别为活体的可能性。反之，如果将第五阈值设置的较低，则更多的人脸检测框将会被确定为是重合的，会增加活体被识别为非活体的可能性。在应用中，第五阈值可以根据注册认证应用的场合进行设置。例如，对于一些利用本公开实施例的方法来进行解锁功能的场合，可以将第五阈值设置的较高些，因为在这些场合中，基本可以保证对象为活体，因此降低活体被识别为非活体的可能性，可以充分保证活体对象被正确识别，由此提高用户的体验。
通过分析对象的动作行为可以确定人脸是否是活体人脸。即在本公开的实施例中，仅通过对多个图像帧中人脸检测框之间的相对位置进行分析就可以确定对象是否是活体，由此能够有效地阻止基于非活体的视频进行解锁的操作，例如可以避免使用对象的照片进行解锁的操作，提高了锁定的安全性。
图6示出了根据本公开实施例的在多个人脸检测框中确定符合重合条件的人脸检测框的过程。如图6所示，在得到第1个图像帧中的人脸检测框F 1与第2个图像帧中的人脸检测框F 2之间的交集区域之后，需要分别计算该交集区域F 12的面积与人脸检测框F 1的面积的第四比率F 12/F 1，以及计算该交集区域F 12的面积与人脸检测框F 2的面积的第四比率F 12/F 2。这里，仍以F 1和F 2表示人脸检测框区域F 1和F 2的面积。然后，需要分 别比较第四比率F 12/F 1和F 12/F 2与第五阈值的关系，只有在第四比率F 12/F 1和F 12/F 2均大于第五阈值的情况下，才确定人脸检测框F 1与人脸检测框F 2符合重合条件。同理计算人脸检测框F 1与人脸检测框F 7的第四比率F 17/F 1和F 17/F 7，如果第四比率F 17/F 1和F 17/F 7均小于或等于第五阈值，则确定人脸检测框F 1与人脸检测框F 7不符合重合条件。
根据实施例，如果第二比率中的一个第二比率小于或等于第二阈值，另一个第二比率大于第二阈值，则确定人脸是非活体人脸。这是一种比较特殊的情况，造成这种情况的原因在于两个人脸检测框的尺寸相差较大。如图6所示，各人脸检测框(例如F 1、F 2、F 7)被示出为具有相同的尺寸。实际中，人脸检测框的尺寸可以彼此不相同，但彼此之间的尺寸相差不大。如果某个人脸检测框的尺寸与其他人脸检测框的尺寸相差较大，则说明该人脸检测框中所包含的人脸可能运动幅度很大，或者该人脸检测框中所包含的人脸可能与其他人脸检测框中所包含的人脸不属于同一个人。因此，这种情况下，可以根据第四比率与第五阈值的比较结果直接确定对象是非活体，不再继续确定其他人脸检测框是否符合重合条件。
图7示出了根据本公开另一实施例的基于视频数据进行注册的装置700的框图。如图7所示，该装置700包括处理器701、存储器702和摄像头703，在存储器702中存储有机器可读指令，处理器701可以执行这些机器可读指令来实现根据本公开实施例的基于视频数据进行人脸注册的方法100。摄像头703可以被配置为获取视频数据，且摄像头703的帧数可以在每秒15～25帧的范围内。
存储器702可以具有以下形式：非易失性或易失性存储器，例如，电可擦除可编程只读存储器(EEPROM)、闪存等。
根据本公开实施例的装置700内部的各种组件可以通过多种器件来实现，这些器件包括但不限于：模拟电路器件、数字电路器件、数字信号处理(DSP)电路、可编程处理器、专用集成电路(ASIC)、现场可编程门阵列(FPGA)、可编程逻辑器件(CPLD)，等等。
图8示出了根据本公开另一实施例的电子白板800的框图。如图8所示，根据本公开实施例的电子白板800包括显示白板801和根据本公开实施例的基于视频数据进行人脸注册的装置802。
根据本公开实施例的电子白板，安装了基于视频数据进行注册的装置，无需人为交互，直接通过视频流截取方式进行人脸注册，采用直接获取视频帧的方式注册更加方便。 根据本公开实施例的电子白板无需手动开关机，一定距离内就可以通过人脸信息直接解锁使用，保密性好。并且，只有预约注册的固定人脸才可以解锁，有效的保护了预约用户在电子白板使用过程中的信息安全。
以上的详细描述通过使用示意图、流程图和/或示例，已经阐述了众多实施例。在这种示意图、流程图和/或示例包含一个或多个功能和/或操作的情况下，本领域技术人员应理解，这种示意图、流程图或示例中的每一功能和/或操作可以通过各种结构、硬件、软件、固件或实质上它们的任意组合来单独和/或共同实现。
虽然已参照几个典型实施例描述了本公开，但应当理解，所用的术语是说明和示例性、而非限制性的术语。由于本公开能够以多种形式具体实施而不脱离公开的精神或实质，所以应当理解，上述实施例不限于任何前述的细节，而应在随附权利要求所限定的精神和范围内广泛地解释，因此落入权利要求或其等效范围内的全部变化和改型都应为随附权利要求所涵盖。

Claims (16)

  1. 一种基于视频数据进行人脸注册的方法，包括：
    接收视频数据；
    从所述视频数据中获取第一图像帧序列，所述第一图像帧序列中的每个图像帧各自包括包含完整人脸特征的人脸检测框；
    根据每个图像帧中的人脸检测框之间的相对位置确定所述图像帧是否达到预设清晰度；
    在确定所述图像帧达到预设清晰度的情况下，基于所述多个人脸检测框的图像信息提取多组人脸特征，并根据所述多组人脸特征确定所述人脸是否表示同一对象；以及
    在确定所述人脸表示同一对象的情况下，根据所述第一图像帧序列注册所述对象。
  2. 根据权利要求1所述的方法，其中，从所述视频数据中获取第一图像帧序列包括：
    按照视频的捕获顺序从所述视频数据中获取多个图像帧；
    基于人脸检测模型确定所述图像帧是否包含人脸；以及
    在确定所述图像帧包含人脸的情况下，在所述多个图像帧中的每个图像帧中确定包含所述人脸的人脸检测框。
  3. 根据权利要求2所述的方法，其中，从所述视频数据中获取第一图像帧序列还包括：
    确定所获取的图像帧是否包含完整人脸特征；
    在所述图像帧包含完整人脸特征的情况下，将所述图像帧存储为第一图像帧序列中的一帧；
    在已存储的第一图像帧序列包括预定帧数的情况下，结束获取图像帧。
  4. 根据权利要求3所述的方法，其中，确定所获取的图像帧是否包含完整人脸特征包括：
    基于人脸姿态检测模型确定所述人脸是否是正脸；
    在确定所述图像帧中所包含的人脸是正脸的情况下，基于人脸遮挡检测模型确定所述人脸是否被遮挡；
    在确定所述图像帧中所包含的人脸未被遮挡的情况下，确定所述图像帧包含完整人脸特征；以及
    否则，确定所述图像帧不包含完整人脸特征。
  5. 根据权利要求1至4中任一项所述的方法，其中，根据每个图像帧中的人脸检测框之间的相对位置确定所述图像帧是否达到预设清晰度包括：
    确定所述第一图像帧序列中的两个图像帧中的人脸检测框的交集区域的面积相对于该两个图像帧中的人脸检测框的并集区域的面积的第一比率；以及
    在所确定的第一比率均大于第一阈值的情况下，确定所述图像帧达到预设清晰度。
  6. 根据权利要求1至4中任一项所述的方法，其中，根据每个图像帧中的人脸检测框之间的相对位置确定所述图像帧是否达到预设清晰度包括：
    确定所述第一图像帧序列中的两个图像帧中的人脸检测框的交集区域的面积相对于该两个图像帧中的人脸检测框的并集区域的面积的第一比率；
    确定大于第一阈值的所述第一比率的数量相对于所述第一比率总的数量的第二比率；以及
    在所述第二比率大于或等于第二阈值的情况下，确定所述图像帧达到预设清晰度。
  7. 根据权利要求5或6所述的方法，其中，根据所述多组人脸特征确定所述人脸是否表示同一对象包括：
    确定所述第一图像帧序列中任意相邻两个图像帧中的人脸特征之间的相似度；以及
    在所确定的相似度均大于第三阈值的情况下，确定所述人脸表示同一对象。
  8. 根据权利要求7所述的方法，其中，所述人脸特征包括人脸特征向量，并且其中，确定所述第一图像帧序列中任意相邻两个图像帧中的人脸特征之间的相似度包括：
    确定所述第一图像帧序列中相邻两个图像帧中的人脸特征向量之间的距离。
  9. 根据权利要求1至8中任一项所述的方法，其中，根据所述第一图像帧序列注册所述对象包括：
    以所述第一图像帧序列中的指定图像帧作为注册数据注册所述对象。
  10. 根据权利要求1所述的方法，还包括：
    将根据所述第一图像帧序列注册所述对象而得到的注册数据存储为人脸库；以及
    基于所述人脸库对所接收的视频数据中的人脸进行识别。
  11. 根据权利要求10所述的方法，其中，基于所述人脸库对所接收的视频数据中的人脸进行识别包括：
    从接收的视频数据中获取第二图像帧序列，所述第二图像帧序列中的每个图像帧各自包括包含完整人脸特征的人脸检测框；
    根据每个图像帧中的人脸检测框之间的相对位置确定所述图像帧是否包括活体人脸；
    在确定所述图像帧包括活体人脸的情况下，基于所述人脸检测框提取人脸特征；以及
    确定所述人脸特征是否与所述人脸库中的注册数据相匹配，以识别所述人脸。
  12. 根据权利要求11所述的方法，其中，根据每个图像帧中的人脸检测框之间的相对位置确定所述图像帧是否包括活体人脸包括：
    在每个图像帧中的人脸检测框中确定符合重合条件的人脸检测框；
    确定所述符合重合条件的人脸检测框的数量相对于所述人脸检测框中全部人脸检测框的数量的第三比率；以及
    在所述第三比率大于或等于第四阈值的情况下，确定所述人脸是非活体人脸；在所述第三比率小于所述第四阈值的情况下，确定所述人脸是活体人脸。
  13. 根据权利要求12所述的方法，其中，在每个图像帧中的人脸检测框中确定符合重合条件的人脸检测框包括：
    确定所述人脸检测框中任意两个人脸检测框之间的交集区域的面积相对于该任意两个人脸检测框中每个人脸检测框的面积的第四比率；
    在所确定的第四比率均大于第五阈值的情况下，确定所述任意两个人脸检测框是符合重合条件的人脸检测框；以及
    在所确定的第四比率均小于第五阈值的情况下，确定所述任意两个人脸检测框是不符合重合条件的人脸检测框。
  14. 根据权利要求13所述的方法，其中，根据每个图像帧中的人脸检测框之间的相对位置确定所述图像帧是否包括活体人脸还包括：
    在所确定的第四比率中的一个第四比率大于所述第五阈值且另一个第四比率小于或等于所述第五阈值的情况下，确定所述人脸是非活体人脸。
  15. 一种基于视频数据进行人脸注册的装置，包括：
    存储器，配置为存储指令；以及
    处理器，配置为执行所述指令，以执行如权利要求1至14中任一项所述的方法。
  16. 一种电子白板，包括如权利要求15所述的装置。
PCT/CN2020/139354 2020-12-25 2020-12-25 基于视频数据进行人脸注册的方法、装置和电子白板 WO2022133993A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
PCT/CN2020/139354 WO2022133993A1 (zh) 2020-12-25 2020-12-25 基于视频数据进行人脸注册的方法、装置和电子白板
CN202080003669.XA CN115053268A (zh) 2020-12-25 2020-12-25 基于视频数据进行人脸注册的方法、装置和电子白板
US17/595,605 US11908235B2 (en) 2020-12-25 2020-12-25 Method and device of registering face based on video data, and electronic whiteboard

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/139354 WO2022133993A1 (zh) 2020-12-25 2020-12-25 基于视频数据进行人脸注册的方法、装置和电子白板

Publications (1)

Publication Number Publication Date
WO2022133993A1 true WO2022133993A1 (zh) 2022-06-30

Family

ID=82158658

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/139354 WO2022133993A1 (zh) 2020-12-25 2020-12-25 基于视频数据进行人脸注册的方法、装置和电子白板

Country Status (3)

Country Link
US (1) US11908235B2 (zh)
CN (1) CN115053268A (zh)
WO (1) WO2022133993A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20230015971A1 (en) * 2021-07-19 2023-01-19 Flawless Holdings Limited Neural network for audio and video dubbing with 3d facial modelling

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105760817A (zh) * 2016-01-28 2016-07-13 深圳泰首智能技术有限公司 用人脸识别认证解锁加密存储空间的方法及装置
CN108171207A (zh) * 2018-01-17 2018-06-15 百度在线网络技术（北京）有限公司 基于视频序列的人脸识别方法和装置
CN110276277A (zh) * 2019-06-03 2019-09-24 罗普特科技集团股份有限公司 用于检测人脸图像的方法和装置
CN110705478A (zh) * 2019-09-30 2020-01-17 腾讯科技（深圳）有限公司 人脸跟踪方法、装置、设备及存储介质
KR102078249B1 (ko) * 2019-06-18 2020-02-17 주식회사 페이스테그 얼굴 인증 방법, 얼굴 인증 장치, 이를 위한 컴퓨터 프로그램

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5476955B2 (ja) * 2009-12-04 2014-04-23 ソニー株式会社 画像処理装置および画像処理方法、並びにプログラム
WO2018111287A1 (en) * 2016-12-15 2018-06-21 Hewlett-Packard Development Company, L.P. Image storage
CN110909695B (zh) * 2017-07-29 2023-08-18 Oppo广东移动通信有限公司 防伪处理方法及相关产品
CN108509896B (zh) * 2018-03-28 2020-10-13 腾讯科技（深圳）有限公司 一种轨迹跟踪方法、装置和存储介质
CN110348270B (zh) * 2018-04-03 2023-06-09 扬智科技股份有限公司 影像物件辨识方法与影像物件辨识系统
TWI671685B (zh) * 2018-09-19 2019-09-11 和碩聯合科技股份有限公司 臉部識別方法與使用此方法的電子裝置
CN110245612A (zh) * 2019-06-14 2019-09-17 百度在线网络技术（北京）有限公司 人脸图像的检测方法和装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105760817A (zh) * 2016-01-28 2016-07-13 深圳泰首智能技术有限公司 用人脸识别认证解锁加密存储空间的方法及装置
CN108171207A (zh) * 2018-01-17 2018-06-15 百度在线网络技术（北京）有限公司 基于视频序列的人脸识别方法和装置
CN110276277A (zh) * 2019-06-03 2019-09-24 罗普特科技集团股份有限公司 用于检测人脸图像的方法和装置
KR102078249B1 (ko) * 2019-06-18 2020-02-17 주식회사 페이스테그 얼굴 인증 방법, 얼굴 인증 장치, 이를 위한 컴퓨터 프로그램
CN110705478A (zh) * 2019-09-30 2020-01-17 腾讯科技（深圳）有限公司 人脸跟踪方法、装置、设备及存储介质

Also Published As

Publication number Publication date
CN115053268A (zh) 2022-09-13
US20220398866A1 (en) 2022-12-15
US11908235B2 (en) 2024-02-20

Similar Documents

Publication Publication Date Title
CN109376684B (zh) 一种人脸关键点检测方法、装置、计算机设备和存储介质
WO2019196308A1 (zh) 人脸识别模型的生成装置、方法及计算机可读存储介质
US9818023B2 (en) Enhanced face detection using depth information
TW201911130A (zh) 一種翻拍影像識別方法及裝置
JP5629803B2 (ja) 画像処理装置、撮像装置、画像処理方法
WO2017096753A1 (zh) 人脸关键点跟踪方法、终端和非易失性计算机可读存储介质
CN109934065B (zh) 一种用于手势识别的方法和装置
US10650260B2 (en) Perspective distortion characteristic based facial image authentication method and storage and processing device thereof
CN112287866A (zh) 一种基于人体关键点的人体动作识别方法及装置
CN112287867B (zh) 一种多摄像头的人体动作识别方法及装置
CN112287868B (zh) 一种人体动作识别方法及装置
CN111931594A (zh) 人脸识别活体检测方法、装置、计算机设备和存储介质
US11315360B2 (en) Live facial recognition system and method
US20220044039A1 (en) Living Body Detection Method and Device
WO2022133993A1 (zh) 基于视频数据进行人脸注册的方法、装置和电子白板
JP2018088049A (ja) 画像処理装置、画像処理方法、及びプログラム
CN110363111B (zh) 基于镜头失真原理的人脸活体检测方法、装置及存储介质
CN109753886B (zh) 一种人脸图像的评价方法、装置及设备
CN109241942B (zh) 图像处理方法、装置、人脸识别设备及存储介质
EP3699865B1 (en) Three-dimensional face shape derivation device, three-dimensional face shape deriving method, and non-transitory computer readable medium
WO2023028947A1 (zh) 掌静脉非接触式三维建模方法、装置及认证方法
JP2020052788A (ja) 画像処理装置及びその方法、プログラム
CN114332981A (zh) 人脸活体检测方法、装置、电子设备及存储介质
CN113221606A (zh) 一种基于ims视频会议登录的人脸识别方法
CN110991307A (zh) 人脸识别的方法、装置、设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20966558

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 20/10/2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20966558

Country of ref document: EP

Kind code of ref document: A1