Search

CN-122024027-A - Digital human video live broadcast detection method and device

CN122024027ACN 122024027 ACN122024027 ACN 122024027ACN-122024027-A

Abstract

The application provides a digital human live video detection method and device, the method comprises the steps of S1, determining three-dimensional coordinates of a plurality of skeleton nodes of a person in live video, S2, calculating displacement vectors of adjacent frames of each skeleton node, S3, calculating first dot product coefficients of two displacement vectors of the same skeleton node at a set time interval, S4, calculating second dot product coefficients of the displacement vector of a selected first skeleton node and the relative displacement vector of a second skeleton node constrained by the topology of the first skeleton node relative to the first skeleton node, and S5, determining whether the person in live video belongs to a digital person according to the first dot product coefficients and/or the second dot product coefficients. The application can rapidly and accurately detect the digital human video.

Inventors

  • WANG HE

Assignees

  • 北京风平智能科技有限公司

Dates

Publication Date
20260512
Application Date
20260205

Claims (10)

  1. 1. A digital human live video detection method, comprising: step S1, determining three-dimensional coordinates of a plurality of skeleton nodes of a person in live video broadcast; S2, calculating displacement vectors of adjacent frames of each skeleton node; s3, calculating first dot product coefficients of two displacement vectors of the same skeleton node at a set time interval; s4, calculating a second dot product coefficient of a displacement vector of the selected first bone node and a relative displacement vector of a second bone node which is topologically constrained by the first bone node relative to the first bone node; And S5, determining whether the person in the live video broadcast belongs to the digital person or not according to the first dot product coefficient and/or the second dot product coefficient.
  2. 2. The method of claim 1, wherein in step S3, the time interval is set to 5-20 video frames.
  3. 3. The digital human live video detection method as claimed in claim 1, wherein step S5 further comprises: And when the duty ratio of any bone node, which is calculated continuously for a plurality of times, of the first dot product coefficient which is smaller than a first set value exceeds a preset value, judging that the person in the live video broadcast is a digital person.
  4. 4. The digital human live video detection method as claimed in claim 1, wherein step S5 further comprises: And when each second dot product coefficient calculated by a plurality of groups of skeleton nodes with topological constraint relation is larger than a second set value, judging that the person in the live video broadcast is a true person, otherwise, judging that the person is a digital person.
  5. 5. The digital human live video detection method as claimed in claim 1, wherein step S5 further comprises: When the index value calculated by the first dot product coefficient and the second dot product coefficient exceeds a third set value, determining that the person in the live video broadcast is a digital person, wherein the index value S is calculated by the following formula: ; Wherein, the 、 The weight coefficients are related to different bone nodes, and are obtained by inquiring a preset table, p is a first dot product coefficient, and q is a second dot product coefficient.
  6. 6. A digital human video live broadcast detection device, comprising: The node coordinate determining module is used for determining three-dimensional coordinates of a plurality of skeleton nodes of the person in the live video broadcast; The displacement vector calculation module is used for calculating the displacement vector of the adjacent frame of each skeleton node; A first coefficient calculation module, configured to calculate first coefficients of two displacement vectors of a same bone node at a set time interval; A second dot product coefficient calculation module, configured to calculate a second dot product coefficient of a displacement vector of a selected first bone node and a relative displacement vector of a second bone node, which is topologically constrained by the first bone node, with respect to the first bone node; and the digital person identification module is used for determining whether the person in the live video broadcast belongs to the digital person according to the first dot product coefficient and/or the second dot product coefficient.
  7. 7. The digital live video detection apparatus of claim 6, wherein the set time interval is 5-20 video frames.
  8. 8. The digital person live video detection apparatus of claim 6, wherein the digital person identification module comprises: and the identification unit is used for judging that the person in the live video broadcast is a digital person when the duty ratio of any bone node in the continuous repeated calculation of the first dot product coefficient smaller than a first set value exceeds a preset value.
  9. 9. The digital person live video detection apparatus of claim 6, wherein the digital person identification module comprises: And the identification unit based on the second dot product coefficients is used for judging that the person in the live video broadcast is a real person when each second dot product coefficient calculated by a plurality of groups of skeleton nodes with topological constraint relations is larger than a second set value, and otherwise, the person is a digital person.
  10. 10. The digital person live video detection apparatus of claim 6, wherein the digital person identification module comprises: The comprehensive identification unit is used for determining that the person in the video live broadcast is a digital person when the index value calculated by the first dot product coefficient and the second dot product coefficient exceeds a third set value, and the index value S is calculated by the following formula: ; Wherein, the 、 The weight coefficients are related to different bone nodes, and are obtained by inquiring a preset table, p is a first dot product coefficient, and q is a second dot product coefficient.

Description

Digital human video live broadcast detection method and device Technical Field The application belongs to the technical field of video processing, and particularly relates to a digital human video live broadcast detection method and device. Background The digital human video is a virtual human video generated by artificial intelligence and computer graphics technology, and is widely applied to various industries such as electronic commerce, advertisement, news broadcasting, training, education and the like. While digital people have their legal use, malicious use in live scenes can cause serious problems. It is therefore necessary to identify the video as a digital person or a real person. Disclosure of Invention In order to solve at least one of the technical problems, the application provides a method and a device for detecting live video of a digital person, which are used for detecting the person in the video by utilizing the principle of inverting materials in the training process of the digital person. In a first aspect of the present application, a digital live video detection method mainly includes: step S1, determining three-dimensional coordinates of a plurality of skeleton nodes of a person in live video broadcast; S2, calculating displacement vectors of adjacent frames of each skeleton node; s3, calculating first dot product coefficients of two displacement vectors of the same skeleton node at a set time interval; s4, calculating a second dot product coefficient of a displacement vector of the selected first bone node and a relative displacement vector of a second bone node which is topologically constrained by the first bone node relative to the first bone node; And S5, determining whether the person in the live video broadcast belongs to the digital person or not according to the first dot product coefficient and/or the second dot product coefficient. Preferably, in step S3, the time interval is set to 5-20 video frames. Preferably, step S5 further includes: And when the duty ratio of any bone node, which is calculated continuously for a plurality of times, of the first dot product coefficient which is smaller than a first set value exceeds a preset value, judging that the person in the live video broadcast is a digital person. Preferably, step S5 further includes: And when each second dot product coefficient calculated by a plurality of groups of skeleton nodes with topological constraint relation is larger than a second set value, judging that the person in the live video broadcast is a true person, otherwise, judging that the person is a digital person. Preferably, step S5 further includes: When the index value calculated by the first dot product coefficient and the second dot product coefficient exceeds a third set value, determining that the person in the live video broadcast is a digital person, wherein the index value S is calculated by the following formula: ; Wherein, the 、The weight coefficients are related to different bone nodes, and are obtained by inquiring a preset table, p is a first dot product coefficient, and q is a second dot product coefficient. The second aspect of the present application provides a digital live video detection device, mainly comprising: The node coordinate determining module is used for determining three-dimensional coordinates of a plurality of skeleton nodes of the person in the live video broadcast; The displacement vector calculation module is used for calculating the displacement vector of the adjacent frame of each skeleton node; A first coefficient calculation module, configured to calculate first coefficients of two displacement vectors of a same bone node at a set time interval; A second dot product coefficient calculation module, configured to calculate a second dot product coefficient of a displacement vector of a selected first bone node and a relative displacement vector of a second bone node, which is topologically constrained by the first bone node, with respect to the first bone node; and the digital person identification module is used for determining whether the person in the live video broadcast belongs to the digital person according to the first dot product coefficient and/or the second dot product coefficient. Preferably, the set time interval is 5-20 video frames. Preferably, the digital person identification module includes: and the identification unit is used for judging that the person in the live video broadcast is a digital person when the duty ratio of any bone node in the continuous repeated calculation of the first dot product coefficient smaller than a first set value exceeds a preset value. Preferably, the digital person identification module includes: And the identification unit based on the second dot product coefficients is used for judging that the person in the live video broadcast is a real person when each second dot product coefficient calculated by a plurality of groups of skeleton nodes with topological constra