WO2021000829A1 - Procédé et appareil d'identification d'informations d'identité multidimensionnelles, dispositif informatique et support de stockage - Google Patents

Procédé et appareil d'identification d'informations d'identité multidimensionnelles, dispositif informatique et support de stockage Download PDF

Info

Publication number
WO2021000829A1
WO2021000829A1 PCT/CN2020/098801 CN2020098801W WO2021000829A1 WO 2021000829 A1 WO2021000829 A1 WO 2021000829A1 CN 2020098801 W CN2020098801 W CN 2020098801W WO 2021000829 A1 WO2021000829 A1 WO 2021000829A1
Authority
WO
WIPO (PCT)
Prior art keywords
preset
identified
identity information
statistical
recognized
Prior art date
Application number
PCT/CN2020/098801
Other languages
English (en)
Chinese (zh)
Inventor
方成银
Original Assignee
平安科技（深圳）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 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021000829A1 publication Critical patent/WO2021000829A1/fr

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • 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
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/56—Extraction of image or video features relating to colour
    • 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
    • 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
    • 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
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/07—Target detection
    • 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
    • G06V40/178—Human faces, e.g. facial parts, sketches or expressions estimating age from face image; using age information for improving recognition
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Definitions

  • This application relates to the field of artificial intelligence technology, in particular to a multi-dimensional identity information recognition method, device, computer equipment and storage medium.
  • the existing applications for target recognition and capture in video basically use the method of face recognition to track the target object.
  • the requirements for scenes such as the posture lighting of the face in the video are high, and the clarity of the video is also required.
  • Very high requirements the inventor realized that in the case of unsatisfactory recognition conditions, due to the small difference in facial features, the equipment requirements are high, and the misrecognition rate is high, which is not conducive to use in complex scenarios, and the recognition speed is not enough fast.
  • the main purpose of this application is to provide a multi-dimensional identification information identification method, device, computer equipment and storage medium, which can quickly identify the identity of the object to be identified in the video and improve the target identification speed in complex scenarios.
  • This application proposes a multi-dimensional identity information identification method, including:
  • the identification information of the object to be identified in the video is identified, and the identification probability of the object to be identified in all categories of the preset multi-dimensional identity information to be identified is obtained; where the preset multi-dimensional identity information to be identified includes N categories of first identity information to be identified in the human body feature dimension and M categories of second identity information to be identified in the clothing dimension; preset multi-dimensional identity information to be identified as the identity information corresponding to the target object, and both M and N Positive integer;
  • the first preset determination table Look up the corresponding first determination result in the first preset determination table according to the recognition probability to determine whether the corresponding type of preset multi-dimensional identity information to be recognized is recognized according to the first determination result; in the first preset determination table, If the types of multi-dimensional identity information to be identified are different, the judgment conditions corresponding to the identification probabilities are not completely the same;
  • This application also proposes a multi-dimensional identity information recognition device, including:
  • the recognition module is used to identify the identity information of the object to be identified in the video according to the target detection algorithm, and obtain the corresponding identification probability of the object to be identified in all categories of preset multi-dimensional identity information to be identified; wherein, the preset to be identified
  • the multi-dimensional identity information includes N categories of first identity information to be identified in the human body feature dimension and M categories of second identity information to be identified in the clothing dimension; the preset multi-dimensional identity information to be identified is the identity information corresponding to the target object, Both M and N are positive integers;
  • the first search module is configured to search for the corresponding first determination result in the first preset determination table according to the recognition probability, so as to determine whether the corresponding type of preset multi-dimensional identity information to be recognized is recognized according to the first determination result;
  • a predetermined judgment table presets the judgment conditions of the recognition probability, where different types of preset multi-dimensional identity information to be recognized have different types, and the judgment conditions corresponding to the recognition probabilities are not completely the same;
  • the second search module is used to perform statistics on a plurality of first determination results according to preset statistical rules, and search for corresponding second determination results in the second preset determination table according to the statistical results to determine whether the object to be identified is the target The objects match; second judgment results corresponding to different statistical results are preset in the second preset judgment table.
  • This application also proposes a computer device, including a memory and a processor, the memory stores a computer program, and the processor executes the computer program to realize the steps of a multi-dimensional identity information identification method:
  • the identification information of the object to be identified in the video is identified, and the identification probability of the object to be identified in all categories of the preset multi-dimensional identity information to be identified is obtained; where the preset multi-dimensional identity information to be identified includes N categories of first identity information to be identified in the human body feature dimension and M categories of second identity information to be identified in the clothing dimension; preset multi-dimensional identity information to be identified as the identity information corresponding to the target object, and both M and N Positive integer;
  • the first preset determination table Look up the corresponding first determination result in the first preset determination table according to the recognition probability to determine whether the corresponding type of preset multi-dimensional identity information to be recognized is recognized according to the first determination result; in the first preset determination table, If the types of multi-dimensional identity information to be identified are different, the judgment conditions corresponding to the identification probabilities are not completely the same;
  • This application also proposes a storage medium on which a computer program is stored.
  • the steps of a multi-dimensional identity information identification method are realized:
  • the identification information of the object to be identified in the video is identified, and the identification probability of the object to be identified in all categories of the preset multi-dimensional identity information to be identified is obtained; where the preset multi-dimensional identity information to be identified includes N categories of first identity information to be identified in the human body feature dimension and M categories of second identity information to be identified in the clothing dimension; preset multi-dimensional identity information to be identified as the identity information corresponding to the target object, and both M and N Positive integer;
  • the first preset determination table Look up the corresponding first determination result in the first preset determination table according to the recognition probability to determine whether the corresponding type of preset multi-dimensional identity information to be recognized is recognized according to the first determination result; in the first preset determination table, If the types of multi-dimensional identity information to be identified are different, the judgment conditions corresponding to the identification probabilities are not completely the same;
  • this application proposes a multi-dimensional identity information recognition method, device, computer equipment and storage medium.
  • the identity information of the object to be identified in the video is identified according to the preset multi-dimensional identity information to be identified and the target detection algorithm is used to obtain the preset multi-dimensional identity information to be identified.
  • the target detection algorithm is used to obtain the preset multi-dimensional identity information to be identified.
  • the recognition probability on the object it is finally determined whether the object to be recognized matches the target object according to the recognition probability.
  • this application uses relatively easy-to-obtain multi-dimensional identity information to be identified to quickly identify objects to be identified, and distinguishes the corresponding weights of different types of identity information to be identified, which improves Recognition speed and accuracy in complex scenes.
  • FIG. 1 is a schematic diagram of the steps of a multi-dimensional identity information identification method in an embodiment of this application;
  • FIG. 2 is a schematic diagram of modules of a multi-dimensional identity information identification device in an embodiment of the application
  • FIG. 3 is a schematic block diagram of modules of a computer device in an embodiment of the application.
  • FIG. 4 is a schematic block diagram of modules of a storage medium in an embodiment of the application.
  • a multi-dimensional identity information identification method which includes the following steps:
  • the identification information of the object to be identified in the video is identified, and the identification probability of the object to be identified in all categories of the preset multi-dimensional identity information to be identified is obtained; among them, the multi-dimensional identity to be identified is preset
  • the information includes N categories of first identification information to be identified in the human body feature dimension and M categories of second identification information to be identified in the clothing dimension; the preset multi-dimensional identification information to be identified is the identity information corresponding to the target object, M and N All are positive integers;
  • S2 Search for the corresponding first determination result in the first preset determination table according to the recognition probability to determine whether the corresponding category of identity information to be recognized is recognized according to the first determination result; in the first preset determination table, preset If the types of multi-dimensional identity information to be identified are different, the judgment conditions corresponding to the identification probability are not completely the same;
  • S3 Perform statistics on multiple first determination results according to preset statistical rules, and look up the corresponding second determination results in the second preset determination table according to the statistical results to determine whether the object to be identified matches the target object;
  • the second judgment results corresponding to different statistical results are preset in the second preset judgment table.
  • the method of this application does not directly recognize the face of the object to be identified when identifying the object to be identified in the video to be identified, but first uses the obtained preset multi-dimensional identity information to be identified as An identity information of the target object in order to identify the target object, and then the target detection algorithm is used to identify the object to be identified, and the identification probabilities corresponding to the identification information of the object to be identified in different dimensions are obtained.
  • the obtained recognition probability is searched for the corresponding determination result in the preset determination table to determine whether the object to be identified matches the acquired identity tag, that is, to determine whether the object to be identified is the target object.
  • the multi-dimensional identity information to be identified mainly includes the human body characteristic dimension and the clothing dimension.
  • the human body characteristic dimension refers to the characteristic information of the human body itself, which includes multiple different types of identity information to be identified, such as gender and age. , Body shape, hair, etc.; while the clothing dimension refers to the clothing of the target object, which also includes a number of different types of identification information to be identified, such as one or more of hats, glasses, tops, pants and shoes .
  • the acquired multi-dimensional identity information to be identified is "male, 35 years old, short hair, red hat, blue shirt, black pants", where "male, 35 years old, short hair” is a human body feature
  • "Red hat, blue top, and black pants” are the three different types of identification information to be identified in the clothing dimension.
  • the object to be identified in the video is performed according to the target detection algorithm.
  • the identification and screening of the multi-dimensional identity information to be identified can obtain the identification probabilities of different types of identity information to be identified in different dimensions of the object to be identified.
  • the gender recognition algorithm based on eigenface, the gender recognition method based on Fisher criterion, and the face based on Adaboost (Adaptive Boosting) + SVM (support vector machine) can be used.
  • Gender classification algorithm, etc. to determine whether the gender of the object to be identified in the video is consistent with the acquired identity information to be identified.
  • a face age estimation algorithm that combines LBP (local binarization mode) and HOG (gradient histogram) features can be used to extract the local statistical features of faces that are closely related to age changes, and use CCA
  • the method of (canonical correlation analysis) is integrated, and finally the age is estimated by the method of SVR (Support Vector Machine Regression) to determine whether the age of the object to be identified in the video is consistent with the obtained identity information to be identified.
  • SVR Small Vector Machine Regression
  • the recognition of age can be divided into age groups, such as "children”, “Juvenile”, “youth”, “middle-aged” and “aged”, etc., transform the regression problem of accurate age recognition into the classification problem of age recognition, improve the recognition speed and reduce the recognition error rate.
  • the RGB model or the HSV model can be used for recognition.
  • the image to be recognized can be converted into an HSV model through cvtColor(imgOriginal, imgHSV, COLOR_BGR2HSV); Then perform color detection, such as using void inRange(InputArray src, InputArray lowerb, InputArray upperb, OutputArray dst); function for color detection, the function of this function is to detect whether each pixel of the src image is between lowerb and upperb, if it is, this pixel is set to 255 , And save it in the dst image, otherwise it is 0; the binary image of the target color can be obtained through the function, and then the binary image is opened to remove the noise, and then the closed operation is used to connect the connected domains.
  • the specific color of the clothing of the object to be identified can be detected, so as to determine whether the clothing color of the object to be identified in the video is consistent with the acquired identity information to be identified. Further, the texture of the image can also be detected to learn the material of the clothes on the object to be identified.
  • the output of the vector machine according to the detection algorithm actually belongs to the classification probability of the object to be recognized, and all are a number. For example, in gender recognition, it belongs to "male” or "female". For two classification problems, the result obtained by the detection algorithm during detection is actually the recognition probability.
  • the recognition probability of a "male” object to be recognized is “0.78", and the recognition probability of a female is "0.22", then the output result is (male , 0.78; Female, 0.22), so it is determined that the gender of the identified object is male; for another example, the recognition probability of the top color of "red” is “0.6”, the recognition probability of "yellow” is “0.1”, and it is “blue”
  • the recognition probability of "color” is "0.08", the recognition probability of "green” is "0.12”, and the recognition probability of "orange” is "0.1”, then the output result is (red, 0.6; yellow, 0.1; blue, 0.08; green, 0.12; yellow, 0.1), so it is determined that the color of the shirt with the identification object is red.
  • various dimensions of the identification information to be identified are identified for the objects to be identified in the video, and the corresponding identification probabilities of the identification information to be identified in each category are obtained.
  • the corresponding first determination result is searched in the first preset determination table according to the identification probability to determine whether the multi-dimensional identification information to be identified is recognized. It is assumed that the judgment condition of the recognition probability is preset in the judgment table, wherein the judgment conditions corresponding to different types of identity information to be identified are different.
  • the identification information to be identified is "male, 35 years old, short hair, red hat, blue top, black pants", and statistics are obtained to obtain the identification probability of an object to be identified in different categories.
  • the preset probability threshold for gender is 0.8, that is, if the recognition probability of gender is greater than or equal to 0.8, the first judgment result is that the to-be-recognized
  • the preset probability threshold for age is 0.65, that is, if the recognition probability of a young age is greater than or equal to 0.65, the first judgment result is the youth of the subject to be identified according to the judgment condition, and other dimensions are different.
  • the multiple first judgment results are counted through the preset statistical rules, and according to the statistical results in the first Second, look up the corresponding second determination result in the preset determination table to determine whether the object to be identified matches the target object.
  • the second judgment results corresponding to different statistical results are preset in the second preset judgment table, and the second judgment results include three different results: "matching", "not matching” and "uncertain”.
  • the preset statistical rule only counts the number of "yes” or "no" in the first determination result, and the specific value of each recognition probability.
  • the first determination result is the category that recognizes the category If the identity information to be identified is "Yes”, if the category of identity information to be identified is not recognized, it is "No". According to the number of the first judgment results in the statistical results as "Yes” and “No", Look up the corresponding second determination result in the second preset determination table. In a specific embodiment, for example, only after the first determination result of all categories of identity information to be identified has passed, that is, "Yes", the determination The identity tag of the target object is recognized, the corresponding second determination result is found in the second preset determination table as "match”, and it is determined that the object to be recognized matches the target object. In another specific embodiment, for example, there are five different types of identification information to be identified.
  • the first determination results of the four different types of identification information to be identified are all "yes” , It is determined that the object to be identified matches the target object; further, other requirements can be made for the identification information to be identified that fails the final judgment result, such as limiting its identification probability to be between "0.3-0.7” It can serve as an auxiliary reference under the premise that the first judgment results of the other four different types of identity information to be identified are passed; for example, it can be set to find the corresponding in the second preset judgment table.
  • the second determination result of is "undefined", and the user can mark the object to be identified as a suspected target object according to the second determination result.
  • the identification information of the object to be identified in the video is identified according to the target detection algorithm, and the identification probability of the object to be identified in all categories of preset multi-dimensional identity information to be identified is obtained.
  • S11 Separate the object to be recognized from the background of the video to be recognized by an image segmentation algorithm
  • S12 Perform part key point detection on the separated object to be recognized, and segment the recognition area of the object to be recognized according to the key point of the part, and the recognition area includes the head, upper body, and lower body;
  • the object to be identified when performing the corresponding identification and detection of the object to be identified in the video to be identified, the object to be identified is first separated from the background of the video to be identified, that is, only the object to be identified is detected, and the object to be identified is not detected.
  • the video background is detected to reduce the amount of detection operations.
  • an image segmentation algorithm can be used to separate the object to be recognized from the background of the video to be recognized, such as fixed threshold segmentation based on threshold segmentation, histogram bimodal method, OTSU method (maximum between-class variance method), etc.; Various edge detection operators of edge detection methods; region growing method, split merging method and watershed segmentation method based on region segmentation; Normalized Cuts algorithm based on graph theory segmentation, Graph Cuts algorithm, Superpixel lattice algorithm, etc.; based on energy functional The segmentation method, wavelet-based segmentation method and neural network-based segmentation method, etc.
  • the recognition area includes the head, upper body and lower body, such as the neck between the head and upper body. It is a key point of a part.
  • the waist between the upper body and the lower body is also a key point of the part.
  • the hands, elbows, shoulders, chest, face, etc. are also key points of the part.
  • the key points of multiple parts are detected by the object to be recognized. Construct, divide the object to be recognized into three recognition areas, including the head, upper body and lower body, so as to subsequently identify the identity information to be recognized in the corresponding dimensions in each recognition area.
  • a haar cascade can be used to detect the head, upper body, and lower body of the object to be identified.
  • the identification area After dividing the object to be identified into different identification areas, identify the corresponding type of identity information to be identified in the identification area, and obtain the identification probability.
  • the recognition probability of "wearing a hat” is greater than or equal to 0.9, then it can be directly determined that the recognition probability of the corresponding "red hat” is that the probability of recognizing "red” is 0.8.
  • the probability calculation is specially limited. The same is true for the recognition of black pants in other recognition areas, such as the recognition area of the lower body, and there is no special restriction when calculating the recognition probability.
  • Step S2 includes:
  • S21 Search for a corresponding determination condition in the first preset determination table according to the corresponding category of the recognition probability; the determination condition includes the preset probability threshold of the recognition probability in the corresponding category;
  • S22 Compare the recognition probability with a preset probability threshold; the preset probability threshold includes a first preset probability threshold and a second preset probability threshold;
  • the determination conditions corresponding to the recognition probability are not completely the same, so the corresponding category according to the recognition probability is first A corresponding determination condition is searched in a predetermined determination table to determine how to determine the recognition probability of a given category, and the determination condition includes a predetermined probability threshold of the recognition probability in the corresponding category.
  • the recognition probability is compared with the found preset probability threshold. If it is higher than the first preset probability threshold, it is determined that the identification information of the category to be identified is recognized; if it is higher than the second preset probability threshold and lower than If the first preset probability threshold is not recognized, it is determined whether the category of identity information to be recognized is uncertain; if it is lower than the second preset probability threshold, it is determined that the category of identity information to be recognized is not recognized.
  • the type of identity information to be identified is different, and the corresponding preset probability threshold will also be different. For example, when performing age recognition on the object to be identified, relatively clear facial images are required for age recognition.
  • the first preset probability threshold can be set lower, for example, 0.7; while the object to be identified is identified by long shorts or whether there is a person wearing glasses During recognition, due to the large difference between trousers and shorts, and the difference between wearing glasses and not wearing glasses, and the requirements for image quality during detection are relatively low, the first preset probability threshold can be set Higher, for example 0.9.
  • the preset probability threshold is lower if the error rate is higher, and the preset probability threshold is higher if the error rate is low, so that the identified object can be treated as quickly as possible Make a judgment to determine whether it is consistent with the acquired identity information to be identified, while minimizing detection errors.
  • the first judgment result is divided into three categories: the identification information to be identified, the identification information to be identified is uncertain, and the identification information to be identified is not identified, instead of the simple binary processing of yes or no, so that It can be handled more flexibly when making comprehensive judgments.
  • the identification information to be identified is uncertain
  • the identification information to be identified is not identified, instead of the simple binary processing of yes or no, so that It can be handled more flexibly when making comprehensive judgments.
  • the first determination result of the four different types of identification information to be identified is that the corresponding category of identification information to be identified is recognized, It is determined that the object to be identified is consistent with the target object; further, other requirements can be made for the remaining identity information for which the first judgment result is not passed, such as limiting the first judgment result to be uncertain or not
  • the identification of the identity information to be identified in the corresponding category can serve as an auxiliary reference under the premise that the first determination result of the other four different categories of identity information to be identified is the identification of the identity information to be identified in the corresponding category.
  • the step S3 of performing statistics on the multiple first determination results according to a preset statistical rule includes:
  • the statistical score is assigned a value of 1; if it is determined that the identification information to be identified is uncertain according to the first determination result, then The statistical score is assigned a value of 0; if the identification information to be identified is not recognized according to the first judgment result, the statistical score is assigned a score of -1;
  • a corresponding statistical score is assigned to different first judgment results, such as score A, score B, and score C.
  • the statistical score is assigned a value of 1 (A); if it is determined according to the first determination result that it is uncertain whether the identity to be identified is identified Information, the statistical score is assigned a value of 0 (B); if the identification information to be identified is not recognized according to the first determination result, the statistical score is assigned a value of -1 (C).
  • the statistical scores of all the first judgment results are obtained, the statistical scores of all the first judgment results are superimposed and calculated to obtain the statistical result.
  • the statistical result is an intuitive score value. The higher the score value, the object to be identified The higher the match with the target audience. Then, according to the score value, the corresponding second determination result is searched in the second preset determination table to determine whether the object to be identified matches the target object.
  • a certain score threshold is preset, and when the statistical result is greater than the preset score threshold, the first object matching the target object is found. 2. Judgment results.
  • the step S32 of superimposing the statistical scores of all the first judgment results to obtain the statistical results includes:
  • W1, W2, and W3 are the statistical weights corresponding to different types of preset multi-dimensional identity information to be identified.
  • the statistical scores of all the first judgment results are superimposed according to the first formula to obtain the statistical results.
  • different types of identification information are to be identified.
  • the determination results have their corresponding statistical weights.
  • the statistical weights in the statistical results of the identification probability are defined as W1, W2, W3, W4,
  • the object to be identified is considered to be consistent with the acquired identity information, otherwise the object to be identified is considered to be the same
  • the identity information does not match.
  • the corresponding statistical weights will be different.
  • the statistical weight can be set lower, for example, 0.2; and when the object to be identified is recognized for long shorts or whether there is recognition of wearing glasses, due to the large difference between trousers and shorts , There is a big difference between wearing glasses and not wearing glasses, and the requirements for image quality during detection are relatively low, so the statistical weight can be set higher, for example, 0.5, a variety of different types of identity information to be identified The sum of its statistical weights is 1.
  • the statistical weight is lower if the error rate is higher, and the statistical weight is higher if the error rate is low.
  • the identity information to be identified minimizes the impact of errors caused by the identity information to be identified with a high error rate, so that accurate judgments can be made as quickly as possible on the object to be identified, and whether it is consistent with the acquired identity information to be identified. At the same time minimize detection errors.
  • the step of obtaining the statistical weights corresponding to different types of preset to-be-identified multi-dimensional identity information in the preset weight distribution table includes:
  • the current environmental factors include one or more of the current temperature, current air quality, current visibility, current geographic location, and height from the ground A combination of factors; among them, the current environmental factors pre-associated with different types of preset multi-dimensional identity information to be identified are not completely the same;
  • S3212 Determine the corresponding statistical weight in the preset weight distribution table according to the preset type of the multi-dimensional identity information to be identified and current environmental factors; set the corresponding preset weight distribution table in the category of the preset multi-dimensional identity information to be identified Multiple threshold ranges of environmental factors, and different threshold ranges of environmental factors correspond to different statistical weights.
  • the device first obtains the pre-associated current environmental factors according to the preset type of the multi-dimensional identity information to be identified, and then according to the preset type of the multi-dimensional identity information to be identified and the current environment
  • the factor determines the corresponding statistical weight in the preset weight distribution table. For example, when the device detects real-time video or detects certain specific videos, it can determine the statistical weights corresponding to different dimensions of identity information through the current environmental factors obtained.
  • the current environmental factors include current temperature, current air quality, current visibility, A combination of one or more environmental factors in the current geographic location and height from the ground. When the device is performing detection and identification, it can obtain the current environmental factors through the network.
  • the acquisition method can be the device actively search through the network, such as current environmental factors such as current temperature, current air quality, current geographic location, etc., or the device through the network Receive user input of current environmental factors.
  • the device can also acquire current environmental factors through sensors. For example, when acquiring the current environmental factor of the height from the ground, the distance between the camera and the ground can be obtained through the measurement of the distance sensor on the camera. Among them, the current environmental factors pre-associated with different types of preset multi-dimensional identity information to be identified are not completely the same.
  • the length of pants is mainly related to the current temperature, so the preset multi-dimensional identity information of the type of pants length to be identified
  • the pre-associated current environmental factors include the current temperature, and the pre-recognized multi-dimensional identity information of whether to wear glasses and hair length is not related to the current temperature, so the pre-associated current environmental factors do not include Current temperature; for example, the pre-recognized multi-dimensional identity information of the shoe color category has little correlation with current environmental factors such as current air quality, current visibility, and current geographic location, but it is related to the height from the ground.
  • the current environmental factors have a large correlation, so the current environmental factors pre-associated include the height from the ground.
  • the corresponding statistical weights are determined in the preset weight distribution table according to the preset types of the multi-dimensional identity information to be identified and the current environmental factors; Multiple environmental factor threshold ranges are set to identify the categories of multi-dimensional identity information, and different environmental factor threshold ranges correspond to different statistical weights. For example, when the current temperature is 0°C, although the recognition error rate is low when the object to be identified is identified by long shorts, since when the temperature is 0°C, people basically wear long pants, and the screening significance is more significant.
  • the identity information to be identified is trousers
  • its statistical weight is automatically reduced, for example, to 0.1; and when the current temperature is 0°C and the identity information to be identified is shorts, there are basically few people at this time. Will wear shorts, shorts have a more significant screening significance, then its statistical weight will be automatically increased, for example, to 0.7, and when the current temperature is 30°C, the statistical weight ratio of trousers and shorts will be reversed. People who wear shorts are more common, while those who wear long trousers are relatively small. The statistical weight of trousers is higher, and the statistical weight of shorts is lower. As for whether there is identity information to be identified for wearing glasses, because of whether they wear glasses It has little relationship with the current temperature, so its statistical weight does not change due to changes in the current temperature.
  • the image definition is generally poor, and the height of the video camera is generally relatively low. High, the shoes are not easy to shoot clearly, so the statistical weight of the recognition result of "blue shoes” is generally low, such as 0.1, but if the camera is relatively close to the ground, such as 1 meter, then In the video, a clearer shoe image can be obtained, and its statistical weight is automatically increased, for example, 0.5. The same is true for other types of current environmental factors.
  • the statistical weight is automatically increased; when the current environmental factors have a negative impact on the identification of a certain dimension of identity information to be identified, its statistical weight is automatically reduced; when the current environmental factors affect the identity of a certain dimension to be identified
  • the identification of information has no impact, the statistical weight is determined according to the preset statistical weight or the statistical weight is evenly distributed.
  • the identification information of the object to be identified in the video is identified according to the target detection algorithm, and the identification probabilities corresponding to all categories of the preset multi-dimensional identity information to be identified are obtained respectively.
  • step S1 it also includes:
  • S01 By receiving an input setting instruction of preset multi-dimensional identity information to be identified or performing identity information identification on a designated target object through a target detection algorithm, the preset multi-dimensional identity information to be identified is obtained.
  • the device when detecting the video to be recognized, it is first necessary to determine the preset multi-dimensional identity information of the target object to be recognized.
  • the device when the user knows how to accurately input the setting instructions or there is no image of the target object for identification, the device can receive the user's input settings of the multi-dimensional identity information to be identified through wired or wireless communication. Instructions to determine the identification information to be detected.
  • the user when the user does not know how to enter the identity information to be identified in the corresponding dimension or the device cannot receive the entered identity information to be identified, the user can designate a specific identification object as the target object, and then the device can pass The target detection algorithm recognizes the identity information of the specific identification object to actively obtain the multi-dimensional identity information to be identified.
  • the device detects and recognizes images specified by the user, such as detecting and recognizing a full-body photo of a specific recognized object, and obtains the gender, age, body type, and clothes of the specific recognized object as a pre-recognition.
  • images specified by the user such as detecting and recognizing a full-body photo of a specific recognized object, and obtains the gender, age, body type, and clothes of the specific recognized object as a pre-recognition.
  • multi-dimensional identity information to be identified and then perform detection and recognition in the video according to the identified preset multi-dimensional identity information to be identified, and find the object to be identified that matches the multi-dimensional identity information to be identified.
  • Dimensional identity information is used to detect people in the video that match the specific identification object in the image.
  • this application also proposes a multi-dimensional identity information recognition device, including:
  • the recognition module 10 is used to identify the identity information of the object to be recognized in the video according to the target detection algorithm, and obtain the corresponding recognition probability of the object to be recognized in all categories of preset multi-dimensional identity information to be recognized; Identifying multi-dimensional identity information includes N categories of first identity information to be identified in the human body feature dimension and M categories of second identity information to be identified in the clothing dimension; preset the multi-dimensional identity information to be identified as the identity information corresponding to the target object , M and N are both positive integers;
  • the first search module 20 is configured to search for the corresponding first determination result in the first preset determination table according to the recognition probability, so as to determine whether the corresponding type of preset multi-dimensional identity information to be recognized is recognized according to the first determination result;
  • the judgment conditions of the recognition probability are preset, wherein, if different types of preset multi-dimensional identity information to be identified have different types, the judgment conditions corresponding to the recognition probabilities are not completely the same;
  • the second search module 30 is configured to perform statistics on a plurality of first determination results according to preset statistical rules, and search for corresponding second determination results in the second preset determination table according to the statistical results, to determine whether the object to be identified is compatible with The target object matches; the second judgment result corresponding to different statistical results is preset in the second preset judgment table.
  • the aforementioned modules 10-30 correspondingly include sub-modules, units or sub-units for performing the subdivision steps of the foregoing multi-dimensional identity information identification method , I won’t repeat it here.
  • this application also proposes a computer device, including a memory 1003 and a processor 1002, the memory 1003 stores a computer program 1004, and the processor 1002 executes the computer program 1004 when implementing any of the above methods, including:
  • the identification information of the object to be identified in the video is identified, and the identification probability of the object to be identified in all categories of the preset multi-dimensional identity information to be identified is obtained;
  • the preset multi-dimensional identity information to be identified includes N categories of first identity information to be identified in the human body feature dimension and M categories of second identity information to be identified in the clothing dimension; preset multi-dimensional identity information to be identified as the identity information corresponding to the target object, and both M and N A positive integer;
  • the recognition probability look up the corresponding first judgment result in the first preset judgment table to determine whether the corresponding type of preset multi-dimensional identity information to be identified is recognized according to the first judgment result; in the first preset judgment In the table, if the types of the preset multi-dimensional identity information to be identified are
  • this application also proposes a computer storage medium 2001.
  • the storage medium 2001 may be a non-volatile storage medium or a volatile storage medium.
  • a computer program 2002 is stored thereon.
  • the steps of any one of the above methods include: identifying the object to be identified in the video according to the target detection algorithm to obtain the object to be identified Recognition probabilities corresponding to all categories of preset multi-dimensional identity information to be recognized; wherein, the preset multi-dimensional identity information to be recognized includes N categories of first identity information to be recognized in the human body feature dimension and M categories in the clothing dimension
  • the second identity information to be identified; the preset multi-dimensional identity information to be identified is the identity information corresponding to the target object, M and N are both positive integers; according to the identification probability, look up the corresponding first judgment result in the first preset judgment table , To determine whether the preset multi-dimensional identity information of the corresponding category is recognized according to the first determination result; in the first preset determination table, if the categories of the preset multi-dimensional identity information to

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Human Computer Interaction (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Image Analysis (AREA)

Abstract

La présente invention a trait au domaine technique de l'intelligence artificielle. L'invention concerne un procédé et un appareil d'identification d'informations d'identité multidimensionnelles, ainsi qu'un dispositif informatique et un support de stockage. Le procédé comprend : pendant le suivi et l'identification d'un objet cible dans une vidéo, après l'acquisition d'informations d'identité multidimensionnelles prédéfinies à identifier de l'objet cible, l'identification d'informations d'identité d'un objet à identifier dans la vidéo selon les informations d'identité multidimensionnelles prédéfinies à identifier et au moyen d'un algorithme de détection cible, de manière à obtenir une probabilité d'identification correspondante desdites informations d'identité multidimensionnelles prédéfinies par rapport à l'objet à identifier ; et enfin, la détermination, selon la probabilité d'identification, de la correspondance ou non de l'objet à identifier à l'objet cible. Selon la présente invention, lorsqu'une condition de reconnaissance faciale n'est pas idéale, l'objet à identifier est rapidement distingué au moyen des informations d'identité multidimensionnelles à identifier, lesdites informations étant obtenues relativement facilement, et différents types d'informations d'identité à identifier sont distingués de manière correspondante au moyen de pondérations, de sorte que la vitesse et la précision d'identification dans un scénario complexe soient améliorées.
PCT/CN2020/098801 2019-07-03 2020-06-29 Procédé et appareil d'identification d'informations d'identité multidimensionnelles, dispositif informatique et support de stockage WO2021000829A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910601411.9 2019-07-03
CN201910601411.9A CN110443137B (zh) 2019-07-03 2019-07-03 多维度身份信息识别方法、装置、计算机设备及存储介质

Publications (1)

Publication Number Publication Date
WO2021000829A1 true WO2021000829A1 (fr) 2021-01-07

Family

ID=68428507

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/098801 WO2021000829A1 (fr) 2019-07-03 2020-06-29 Procédé et appareil d'identification d'informations d'identité multidimensionnelles, dispositif informatique et support de stockage

Country Status (2)

Country Link
CN (1) CN110443137B (fr)
WO (1) WO2021000829A1 (fr)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113591539A (zh) * 2021-06-01 2021-11-02 中国电子科技集团公司第三研究所 一种目标识别方法、装置及可读存储介质
CN117558035A (zh) * 2024-01-11 2024-02-13 中国刑事警察学院 一种基于图像技术的人物身份识别系统及方法

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110443137B (zh) * 2019-07-03 2023-07-25 平安科技（深圳）有限公司 多维度身份信息识别方法、装置、计算机设备及存储介质
WO2021098486A1 (fr) * 2019-11-22 2021-05-27 重庆海尔洗衣机有限公司 Procédé de traitement de reconnaissance de couleur, appareil, dispositif et support de stockage
CN111222856A (zh) * 2020-01-15 2020-06-02 深信服科技股份有限公司 一种邮件识别方法、装置、设备及存储介质
CN112966142A (zh) * 2021-04-20 2021-06-15 南京云玑信息科技有限公司 一种准确率与识别速度提高的优化方法
CN113343934A (zh) * 2021-07-14 2021-09-03 华东理工大学 一种基于局部图像差异分析的手部饰品检测方法
CN117421729B (zh) * 2023-12-18 2024-04-26 湖南森鹰科技有限公司 一种自动化程序攻击检测方法、装置、系统及介质
CN117934158A (zh) * 2024-03-13 2024-04-26 湖南三湘银行股份有限公司 一种基于rpa的信贷资料自动审查方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2008191760A (ja) * 2007-02-01 2008-08-21 Toyota Central R&D Labs Inc 対象物検出装置、対象物検出方法、及びプログラム
CN106503687A (zh) * 2016-11-09 2017-03-15 合肥工业大学 融合人脸多角度特征的监控视频人物身份识别系统及其方法
JP2018120644A (ja) * 2018-05-10 2018-08-02 シャープ株式会社 識別装置、識別方法およびプログラム
CN109117801A (zh) * 2018-08-20 2019-01-01 深圳壹账通智能科技有限公司 人脸识别的方法、装置、终端及计算机可读存储介质
CN110443137A (zh) * 2019-07-03 2019-11-12 平安科技（深圳）有限公司 多维度身份信息识别方法、装置、计算机设备及存储介质

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8488023B2 (en) * 2009-05-20 2013-07-16 DigitalOptics Corporation Europe Limited Identifying facial expressions in acquired digital images
US8107726B2 (en) * 2008-06-18 2012-01-31 Samsung Electronics Co., Ltd. System and method for class-specific object segmentation of image data
CN107066983B (zh) * 2017-04-20 2022-08-09 腾讯科技（上海）有限公司 一种身份验证方法及装置
CN109948481B (zh) * 2019-03-07 2024-02-02 惠州学院 一种基于窄带射频链路采样的无源人体识别方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2008191760A (ja) * 2007-02-01 2008-08-21 Toyota Central R&D Labs Inc 対象物検出装置、対象物検出方法、及びプログラム
CN106503687A (zh) * 2016-11-09 2017-03-15 合肥工业大学 融合人脸多角度特征的监控视频人物身份识别系统及其方法
JP2018120644A (ja) * 2018-05-10 2018-08-02 シャープ株式会社 識別装置、識別方法およびプログラム
CN109117801A (zh) * 2018-08-20 2019-01-01 深圳壹账通智能科技有限公司 人脸识别的方法、装置、终端及计算机可读存储介质
CN110443137A (zh) * 2019-07-03 2019-11-12 平安科技（深圳）有限公司 多维度身份信息识别方法、装置、计算机设备及存储介质

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113591539A (zh) * 2021-06-01 2021-11-02 中国电子科技集团公司第三研究所 一种目标识别方法、装置及可读存储介质
CN113591539B (zh) * 2021-06-01 2024-04-16 中国电子科技集团公司第三研究所 一种目标识别方法、装置及可读存储介质
CN117558035A (zh) * 2024-01-11 2024-02-13 中国刑事警察学院 一种基于图像技术的人物身份识别系统及方法
CN117558035B (zh) * 2024-01-11 2024-04-16 中国刑事警察学院 一种基于图像技术的人物身份识别系统及方法

Also Published As

Publication number Publication date
CN110443137A (zh) 2019-11-12
CN110443137B (zh) 2023-07-25

Similar Documents

Publication Publication Date Title
WO2021000829A1 (fr) Procédé et appareil d'identification d'informations d'identité multidimensionnelles, dispositif informatique et support de stockage
CN109558810B (zh) 基于部位分割与融合目标人物识别方法
CN102902959B (zh) 基于二代身份证存储证件照的人脸识别方法及系统
Günther et al. Unconstrained face detection and open-set face recognition challenge
CN102194108B (zh) 一种聚类线性鉴别分析特征选择的笑脸表情识别方法
Aiping et al. Face detection technology based on skin color segmentation and template matching
US9607211B2 (en) Face recognition apparatus, system and method for managing users based on user grouping
CN106503638B (zh) 用于颜色识别的图像处理、车辆颜色识别方法和系统
CN105205437B (zh) 基于头部轮廓验证的侧脸检测方法及装置
WO2019153175A1 (fr) Système et procédé de reconnaissance de visage dissimulé sur la base d'un apprentissage machine, et support de stockage
CN106599880A (zh) 一种面向无人监考的同人判别方法
WO2013075295A1 (fr) Procédé d'identification de vêtements et système pour une vidéo à faible résolution
Hu et al. Fast face detection based on skin color segmentation using single chrominance Cr
CN106980864B (zh) 一种基于支持样本间接式的行人再识别方法
Das et al. Human face detection in color images using HSV color histogram and WLD
Varma et al. Human skin detection using histogram processing and gaussian mixture model based on color spaces
CN106407878B (zh) 基于多分类器的人脸检测方法及装置
Nguyen et al. Reliable detection of eye features and eyes in color facial images using ternary eye-verifier
KR101711328B1 (ko) 영상 촬영 장치를 통하여 획득한 영상 이미지에서 신장 대비 머리 높이의 비를 이용하여 어린이와 성인을 구분하는 방법
Moi et al. A unified approach for unconstrained off-angle iris recognition
WO2023093241A1 (fr) Procédé et appareil de ré-identification de piéton et support de stockage
CN105243380A (zh) 基于选择性中值滤波和pca结合的单人脸图像识别方法
CN109766860A (zh) 基于改进的Adaboost算法的人脸检测方法
KR101851823B1 (ko) 영상 이미지에서 신장 대비 머리 높이의 비를 이용하여 어린이와 성인을 구분하는 방법
Sheng et al. Person re-identification via learning visual similarity on corresponding patch pairs

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: 20834326

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20834326

Country of ref document: EP

Kind code of ref document: A1