Search

CN-116012911-B - Repeated open-broadcast event detection method and device, equipment and medium thereof

CN116012911BCN 116012911 BCN116012911 BCN 116012911BCN-116012911-B

Abstract

The application discloses a repeated open-play event detection method, a device, equipment and a medium thereof, wherein the method comprises the steps of obtaining a target face image in a live video stream of a target user; the method comprises the steps of carrying out semantic matching on the face feature vector of the target face image and the face feature set of the stock users in a user feature library to obtain semantic similarity corresponding to a plurality of face feature vectors in the face feature set of each stock user, judging whether the highest semantic similarity corresponding to each stock user exceeds a preset threshold, and determining that repeated play events exist between the stock users exceeding the preset threshold and the target users. The application accurately identifies repeated broadcasting events based on semantic matching between the target face image of the target user and a plurality of face images of each stock user, effectively protects the network live broadcasting order and can ensure the live broadcasting environment to have social fairness.

Inventors

  • ZHENG KANGYUAN

Assignees

  • 广州方硅信息技术有限公司

Dates

Publication Date
20260512
Application Date
20221227

Claims (10)

  1. 1. A method for detecting repeated open events, comprising the steps of: Acquiring a target face image in a live video stream of a target user; Carrying out semantic matching on the face feature vectors of the target face image and the face feature sets of the stock users in the user feature library to obtain semantic similarity corresponding to a plurality of face feature vectors in the face feature sets of each stock user; Judging whether the highest semantic similarity corresponding to each stock user exceeds a preset threshold value, and determining that repeated broadcasting events exist between the stock users exceeding the preset threshold value and the target users; judging whether the highest semantic similarity corresponding to each stock user exceeds a preset threshold value, and determining that repeated multicast events exist between the stock users exceeding the preset threshold value and the target users, wherein the method comprises the following steps: judging whether the highest semantic similarity corresponding to each stock user exceeds a preset threshold, and when the highest semantic similarity of all the stock users in the user feature library does not exceed the preset threshold, confirming the target user as a new broadcast user; When the candidate stock users with the highest semantic similarity exceeding the preset threshold exist, judging whether the feature identifiers of the candidate stock users and the target users are the same, and when the feature identifiers are the same, confirming the target users as historical users; and when the characteristic identifiers of the candidate stock users and the target users are different, determining that repeated play events exist between the target users and the candidate stock users, and implementing live broadcast authority constraint control on the target users.
  2. 2. The repeated open event detection method according to claim 1, wherein after identifying the target user as a new broadcast user, comprising: Acquiring image frames of part of live video streams of the new broadcasting users, wherein the part of the image frames carry face contents, and extracting face images in the image frames to form an image set to be put in storage; Clustering based on the face feature vectors of the face images in the image set to be put in storage, determining the largest cluster class from a plurality of cluster classes obtained by clustering, and taking the largest cluster class as the face feature set of the new-cast user; and adding the face feature set of the new broadcasting user to the user feature library.
  3. 3. The repeated multicast event detection method according to claim 2, wherein adding the face feature set of the new multicast user to the user feature library comprises: based on the face feature sets of all the new broadcasting users, the semantic similarity is calculated for the face feature vectors in the face feature sets of every two new broadcasting users in a crossing way; comparing whether the highest semantic similarity between every two new broadcasting users exceeds a preset threshold value, and when the highest semantic similarity exceeds the preset threshold value, confirming that repeated broadcasting events exist between the two new broadcasting users; and adding the face feature set of the new broadcasting user which does not have repeated broadcasting events with other new broadcasting users into the user feature library.
  4. 4. The repeated play event detection method according to claim 1, wherein the performing live-broadcast authority constraint control on the target user includes: sending an alarm message to the target user and the candidate stock users forming repeated broadcasting events with the target user; implementing live broadcast authority constraint control on the target user and/or a live broadcast room of candidate stock users forming repeated play events with the target user, wherein the live broadcast authority constraint control comprises any one or more of limiting the use right of a specific function of the live broadcast room, limiting the user flow of the live broadcast room and prohibiting the play of the live broadcast room; and opening application entries of functions of the mutual merging living broadcast room to the target user and the candidate stock users.
  5. 5. The repeated open event detection method according to any one of claims 1 to 4, wherein before acquiring the target face image in the live video stream of the target user, comprising: Acquiring a live user identification list, and calling a historical video stream of a corresponding stock user according to the characteristic identification of each stock user in the list; Performing playback inquiry on the historical video streams of all stock users, inquiring a plurality of image frames carrying face contents, and extracting face images in the image frames of all stock users to form a corresponding image set to be put in storage; Clustering based on the face feature vectors of the image frames in the image set to be put in storage of each stock user, determining the maximum cluster class corresponding to each stock user from a plurality of cluster classes obtained by clustering, and taking the maximum cluster class as the face feature set of the corresponding stock user; and adding the face feature set to the user feature library.
  6. 6. The method for detecting repeated open events according to claim 5, wherein performing playback query on the historical video streams of the respective stock users, querying out a plurality of image frames carrying face contents therein, extracting face images in the respective image frames of the respective stock users to form a corresponding image set to be put in storage, comprises: Performing playback query on the historical video stream of each stock user, performing face detection on the played back image frames, and determining the image frames containing the face content; Extracting corresponding face images from each image frame of each stock user, which contains face content; and storing the face images of each stock user in the image set to be put in the storage of the corresponding stock user after carrying out posture correction.
  7. 7. A repeat-of-play event detection apparatus, comprising: the face image acquisition module is used for acquiring a target face image in a live video stream of a target user; The image similarity matching module is used for carrying out semantic matching on the face feature vectors of the target face image and the face feature sets of the stock users in the user feature library to obtain semantic similarity corresponding to a plurality of face feature vectors in the face feature sets of each stock user; The repeated multicast identification module is used for judging whether the highest semantic similarity corresponding to each stock user exceeds a preset threshold value or not, and determining that repeated multicast events exist between the stock users exceeding the preset threshold value and the target users; the repeated open-sowing identification module comprises: The new broadcasting user identification unit is used for judging whether the highest semantic similarity corresponding to each stock user exceeds a preset threshold, and when the highest semantic similarity of all the stock users in the user feature library does not exceed the preset threshold, the target user is identified as the new broadcasting user; a history user identifying unit, configured to determine whether the feature identifiers of the candidate stock user and the target user are the same when there is a candidate stock user whose highest semantic similarity exceeds the preset threshold, and when the feature identifiers are the same, identify the target user as a history user; and the repeated play confirmation unit is used for determining that repeated play events exist between the target user and the candidate stock user when the characteristic identifiers of the candidate stock user and the target user are different, and implementing live broadcast authority constraint control on the target user.
  8. 8. The repeated open-air event detection device according to claim 7, comprising a new-air image acquisition module configured to acquire image frames with face content carried by a part of live video streams of the new-air user, extract face images in the image frames to form an image set to be put in storage, a new-air feature optimization module configured to cluster based on face feature vectors of the face images in the image set to be put in storage, determine a maximum cluster class from a plurality of clusters obtained by clustering, and use the maximum cluster class as the face feature set of the new-air user, and a new-air feature put module configured to add the face feature set of the new-air user to the user feature library.
  9. 9. An electronic device comprising a central processor and a memory, characterized in that the central processor is arranged to invoke a computer program stored in the memory for performing the steps of the method according to any of claims 1 to 6.
  10. 10. A computer-readable storage medium, characterized in that it stores in the form of computer-readable instructions a computer program implemented according to the method of any one of claims 1 to 6, which, when invoked by a computer, performs the steps comprised by the corresponding method.

Description

Repeated open-broadcast event detection method and device, equipment and medium thereof Technical Field The present application relates to the field of network live broadcast technologies, and in particular, to a method for detecting a repeated broadcast event, a corresponding device, an electronic device, and a computer readable storage medium. Background In a network live broadcast scene, a host user pushes video streams to a live broadcast room, so that application purposes of talent exhibition, information sharing, knowledge education and the like are realized, the host user participates in social labor through the activities to obtain benefits, and overall social benefits are promoted. Some anchor users strive for higher personal benefits and may repeatedly open by registering multiple user identities, resulting in an out-of-order live environment. In view of the above, the method checks repeated on-demand events, maintains the live order of the network live broadcast, ensures that the users of the anchor can have higher social fairness acquisition sense, and has positive significance. In a conventional implementation manner of repeated multicast event identification, feature clustering is performed on a user account, features of the information are extracted according to user registration information or information related to personal identities of other users, such as a user face image, and then clustering is performed to determine the same user group forming a community relationship, so that identification of a suspected repeated multicast relationship is realized. The method is more suitable for identifying the user group concerned by the collaborative brushing of a plurality of user accounts, and individual phenomena are difficult to identify by being accurate to the level of real person information of specific anchor users, so that the method has limited functions in solving the problem of repeated multicast event identification. In more detailed embodiments of repeated broadcast event identification, facial information of different users is obtained, and similarity between the facial information is compared, so that identification of suspected repeated broadcast events is realized. The disadvantage of this approach is mainly that the reliability of the quality of the data source of the face information is not considered, resulting in that even if high similarity can be obtained as a recognition basis, the corresponding face information is not accurate enough due to the general quality of the data source itself, resulting in inaccurate recognition results. Disclosure of Invention It is therefore a primary object of the present application to solve at least one of the above problems and provide a method for detecting repeated open events, and a corresponding apparatus, electronic device and computer readable storage medium thereof. In order to meet the purposes of the application, the application adopts the following technical scheme: The application provides a repeated open-play event detection method which is suitable for one of the purposes of the application, and comprises the following steps: Acquiring a target face image in a live video stream of a target user; Carrying out semantic matching on the face feature vectors of the target face image and the face feature sets of the stock users in the user feature library to obtain semantic similarity corresponding to a plurality of face feature vectors in the face feature sets of each stock user; judging whether the highest semantic similarity corresponding to each stock user exceeds a preset threshold value, and determining that repeated broadcasting events exist between the stock users exceeding the preset threshold value and the target users. Optionally, determining whether the highest semantic similarity corresponding to each stock user exceeds a preset threshold, and determining that a repeated multicast event exists between the stock user exceeding the preset threshold and the target user includes: judging whether the highest semantic similarity corresponding to each stock user exceeds a preset threshold, and when the highest semantic similarity of all the stock users in the user feature library does not exceed the preset threshold, confirming the target user as a new broadcast user; When the candidate stock users with the highest semantic similarity exceeding the preset threshold exist, judging whether the feature identifiers of the candidate stock users and the target users are the same, and when the feature identifiers are the same, confirming the target users as historical users; and when the characteristic identifiers of the candidate stock users and the target users are different, determining that repeated play events exist between the target users and the candidate stock users, and implementing live broadcast authority constraint control on the target users. Optionally, after identifying the target user as the new broadcast user, the