Quick capturing method, device and equipment for video live broadcast violation and medium

Abstract

The invention relates to a video live broadcast violation rapid capturing method and device, equipment and a medium. According to the method, violation auditing, high-risk violation preferential pushing auditing, reuse of manual processing results of repeated violation pictures and support playback, positioning and viewing are carried out through man-machine combination, efficient and accurate capturing and processing of video live broadcast violation are achieved, and the safety of live broadcast platform content is improved.

Description

Technical Field

[0001] The present application relates to the technical field of video data processing in mobile Internet, and in particular to a method, device, equipment and medium for quickly capturing violations in live video broadcasts. Background Art

[0002] With the rapid development of mobile Internet, online video live streaming has become one of the mainstream content dissemination forms, and the dissemination scope of Internet resources has also expanded significantly. However, the openness and real-time nature of live streaming content has greatly increased the probability of illegal behavior. Rapid capture and efficient review of illegal content in live streaming has become an urgent problem to be solved.

[0003] The existing technology usually adopts a method of taking fixed-frequency screenshots of the live stream, inputting the screenshot frames into a pre-built machine review model, and using the model to evaluate the risk level of the screenshot frames to complete the capture and review of illegal content. However, this method of determining the risk of live streaming through a machine review model is relatively simple, and it is easy to make misjudgments in the face of ambiguous screenshot frames, and it is difficult to fully cover the illegal content during the live streaming process. In addition, the screenshot frames are generally captured at a fixed frequency, and there is also a certain risk of missed judgments, and violations may not be completely captured by the screenshot frames. Moreover, since the live streaming scenes are generally relatively fixed, the continuous push of repeated illegal images will generate a large amount of invalid data, increasing the workload of manual review. In addition, the existing methods are inefficient in capturing violations in high-risk live streaming rooms, resulting in untimely responses to violations, and prone to missed reviews or delayed processing, posing a hidden danger to the safe and compliant dissemination of live streaming content. Summary of the invention

[0004] Based on this, it is necessary to provide a method, device, equipment and medium for quickly capturing violations of live video broadcasts in response to the above-mentioned technical problems, so as to achieve efficient and accurate capture and processing of violations of live video broadcasts and improve the security of the content of live broadcast platforms.

[0005] A method for quickly capturing violations of live video broadcasting, the method comprising: Get the live stream address after the user's live broadcast room starts broadcasting, and obtain the screenshot frames and recorded broadcast files during the user's live broadcast by capturing and recording the live stream; The screenshot frames are input into the multi-dimensional machine review model for violation review, and the live broadcast room where the screenshot frames are located is divided into high, medium and low risks according to the machine review results; the machine review results include violation review marks and violation review confidence, and the violation review marks include violation labels and no violation labels; In the review hall, the live broadcast rooms are sorted and displayed in order of risk level from high to low, and auditors are assigned to conduct manual review; Among them, for medium- and low-risk live broadcast rooms, all screenshot frames, real-time images and recorded broadcast files of the live broadcast room are provided to the auditors for review through the live broadcast details page, and the screenshot frames with violation labels are displayed first for review. If the manual review result is that there is a violation, the corresponding review and processing will be carried out according to the violation type of the screenshot frame; if the manual review result is that there is no violation, the violation label of the screenshot frame will be removed, and the risk level of the live broadcast room and the display order of the live broadcast room in the review hall will be lowered; For high-risk live broadcast rooms, screenshot frames with violation labels and high audit confidence are extracted to the high-risk exclusive task library, and exclusive auditors are arranged to perform corresponding audit processing according to the violation type, and the manual processing results of the screenshot frames are stored. When new screenshot frames with violation labels and high audit confidence are generated in the same high-risk live broadcast room, the similarity between the new and old screenshot frames is compared. If similar screenshot frames exist, the manual processing results of the similar screenshot frames are read for reuse; otherwise, the new screenshot frames are extracted to the high-risk exclusive task library for processing.

[0006] In one embodiment, the live streaming address of the user's live streaming room is obtained after the live streaming is started, and the screenshot frames and recorded broadcast files of the user's live streaming are obtained by capturing and recording the live streaming, including: Obtain the user live stream address after the user's live broadcast room starts broadcasting, perform frame capture on the live stream at a fixed frequency, save each frame of the acquired image as a screenshot frame, and number and name the screenshot frames in the form of self-incrementing screenshot numbers to obtain continuous screenshot frames during the user's live broadcast; The live stream is recorded into multiple slice files in ts format, and the slice files are incrementally written into the recorded broadcast file in m3u8 format according to the generation time sequence of the slice files to obtain the recorded broadcast file during the user's live broadcast process.

[0007] In one embodiment, the method further comprises: During the manual review process, for any screenshot frame suspected of violating regulations in any live broadcast room, the time location of the screenshot frame is calculated based on the screenshot frame number and the fixed frequency of the screenshot frame, and the ts format slice file containing the screenshot frame time location in the m3u8 format recording file of the same live broadcast room is retrieved for playback and review to further confirm whether the screenshot frame suspected of violating regulations has violated regulations.

[0008] In one embodiment, the screenshot frame is input into a multi-dimensional machine review model for violation review, and the live broadcast room where the screenshot frame is located is divided into high, medium and low risks according to the machine review results, including: The screenshot frame is input into the multi-dimensional machine review model for violation review, and the violation review mark and violation review confidence of the screenshot frame are output; among them, the multi-dimensional machine review model involves violation identification of pornographic, political, violent and vulgar content; For the screenshot frames that have no violation labels, the live broadcast room where the screenshot frames are located is determined to be a low-risk live broadcast room; For screenshot frames that hit violation labels and whose violation review confidence is in the low confidence interval, the live broadcast room where the screenshot frame is located is determined to be a medium-risk live broadcast room; For screenshot frames that hit the violation label and whose violation review confidence is within the high confidence interval, the live broadcast room where the screenshot frame is located is determined to be a high-risk live broadcast room.

[0009] In one embodiment, the calculation logic of the high and low confidence intervals determined by the multi-dimensional machine review model is: The screenshot frame data set to be reviewed is sent to the auditor and the multi-dimensional machine review model for violation review at the same time, and the manual review results are compared to determine whether they are consistent with the machine review results. If they are consistent, the machine review results are considered valid; otherwise, the machine review results are considered invalid; Count all the machine review results of the screenshot frame data set to be reviewed output by the multi-dimensional machine review model, and group them into intervals according to the violation review confidence scores of all machine review results, obtain the total number of machine review results in each confidence interval, and distinguish between valid and invalid machine review results; The violation review efficiency of each confidence interval is obtained by calculating the number of valid machine review results and the total number of machine review results in each confidence interval. The confidence interval in which the violation review efficiency reaches the preset threshold is set as the high confidence interval; otherwise, the confidence interval in which the violation review efficiency is lower than the preset threshold is set as the low confidence interval.

[0010] In one embodiment, the manual processing results of the screenshot frames are stored. When a new screenshot frame with a violation label and a high audit confidence is generated in the same high-risk live broadcast room, by comparing the similarity between the new and old screenshot frames, if there are similar screenshot frames, the manual processing results of the similar screenshot frames are read for reuse; otherwise, the new screenshot frame is extracted to the high-risk exclusive task library for processing, including: The manual processing results and hash values ​​of the screenshot frames that hit the violation label and have high audit confidence are stored. When a new screenshot frame that hits the violation label and has high audit confidence is generated in the same high-risk live broadcast room, the hash value of the new screenshot frame is compared with the hash value of the screenshot frame stored in the high-risk processing result table to see if they are close. If the hash values ​​are close, read and reuse the manual processing results of the screenshot frames with close hash values ​​to the new screenshot frame for processing; otherwise, extract the new screenshot frame to the high-risk exclusive task library for review and processing; among them, the manual processing results include pass, warning and blacklist.

[0011] A device for quickly capturing violations of live video broadcast, the device comprising: The frame capture and recording module is used to obtain the live stream address after the user's live broadcast room starts broadcasting, and obtain the screenshot frames and recorded broadcast files during the user's live broadcast by capturing and recording the live stream; The machine review module is used to input the screenshot frame into the multi-dimensional machine review model for violation review, and classify the live broadcast room where the screenshot frame is located into high, medium and low risks according to the machine review results; the machine review results include violation review marks and violation review confidence, and the violation review marks include violation labels and no violation labels; The manual review module is used to sort and display the live broadcast rooms in the review hall in order of risk level from high to low, and assign auditors to conduct manual review; among them, for medium and low-risk live broadcast rooms, all screenshot frames, real-time images and recorded broadcast files of the live broadcast room are provided to auditors for review through the live broadcast details page, and the screenshot frames with violation labels are displayed first for review. If the manual review result is that there is a violation, the corresponding review and processing will be carried out for the violation type of the screenshot frame; if the manual review result is that there is no violation, the violation label of the screenshot frame will be removed and the risk level of the live broadcast room will be reduced. and the order in which the live broadcast room is displayed in the review hall; for high-risk live broadcast rooms, extract the screenshot frames with violation labels and high review confidence to the high-risk exclusive task library, arrange exclusive reviewers to perform corresponding review processing according to the violation type, and store the manual processing results of the screenshot frames. When new screenshot frames with violation labels and high review confidence are generated in the same high-risk live broadcast room, by comparing the similarity between the new and old screenshot frames, if similar screenshot frames exist, read the manual processing results of the similar screenshot frames for reuse; otherwise, extract the new screenshot frames to the high-risk exclusive task library for processing.

[0012] In one of the embodiments, the device also includes: a playback positioning module, which is used to obtain the screenshot frame time positioning based on the screenshot frame number and the fixed frequency of the screenshot frame for any screenshot frame suspected of violating the regulations in any live broadcast room during the manual review process, and to retrieve the ts format slice file containing the screenshot frame time positioning in the m3u8 format recording file of the same live broadcast room for playback and viewing, so as to further confirm whether the screenshot frame suspected of violating the regulations has violated the regulations.

[0013] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented: Get the live stream address after the user's live broadcast room starts broadcasting, and obtain the screenshot frames and recorded broadcast files during the user's live broadcast by capturing and recording the live stream; The screenshot frames are input into the multi-dimensional machine review model for violation review, and the live broadcast room where the screenshot frames are located is divided into high, medium and low risks according to the machine review results; the machine review results include violation review marks and violation review confidence, and the violation review marks include violation labels and no violation labels; In the review hall, the live broadcast rooms are sorted and displayed in order of risk level from high to low, and auditors are assigned to conduct manual review; Among them, for medium- and low-risk live broadcast rooms, all screenshot frames, real-time images and recorded broadcast files of the live broadcast room are provided to the auditors for review through the live broadcast details page, and the screenshot frames with violation labels are displayed first for review. If the manual review result is that there is a violation, the corresponding review and processing will be carried out according to the violation type of the screenshot frame; if the manual review result is that there is no violation, the violation label of the screenshot frame will be removed, and the risk level of the live broadcast room and the display order of the live broadcast room in the review hall will be lowered; For high-risk live broadcast rooms, screenshot frames with violation labels and high audit confidence are extracted to the high-risk exclusive task library, and exclusive auditors are arranged to perform corresponding audit processing according to the violation type, and the manual processing results of the screenshot frames are stored. When new screenshot frames with violation labels and high audit confidence are generated in the same high-risk live broadcast room, the similarity between the new and old screenshot frames is compared. If similar screenshot frames exist, the manual processing results of the similar screenshot frames are read for reuse; otherwise, the new screenshot frames are extracted to the high-risk exclusive task library for processing.

[0014] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps: Get the live stream address after the user's live broadcast room starts broadcasting, and obtain the screenshot frames and recorded broadcast files during the user's live broadcast by capturing and recording the live stream; The screenshot frames are input into the multi-dimensional machine review model for violation review, and the live broadcast room where the screenshot frames are located is divided into high, medium and low risks according to the machine review results; the machine review results include violation review marks and violation review confidence, and the violation review marks include violation labels and no violation labels; In the review hall, the live broadcast rooms are sorted and displayed in order of risk level from high to low, and auditors are assigned to conduct manual review; Among them, for medium- and low-risk live broadcast rooms, all screenshot frames, real-time images and recorded broadcast files of the live broadcast room are provided to the auditors for review through the live broadcast details page, and the screenshot frames with violation labels are displayed first for review. If the manual review result is that there is a violation, the corresponding review and processing will be carried out according to the violation type of the screenshot frame; if the manual review result is that there is no violation, the violation label of the screenshot frame will be removed, and the risk level of the live broadcast room and the display order of the live broadcast room in the review hall will be lowered; For high-risk live broadcast rooms, screenshot frames with violation labels and high audit confidence are extracted to the high-risk exclusive task library, and exclusive auditors are arranged to perform corresponding audit processing according to the violation type, and the manual processing results of the screenshot frames are stored. When new screenshot frames with violation labels and high audit confidence are generated in the same high-risk live broadcast room, the similarity between the new and old screenshot frames is compared. If similar screenshot frames exist, the manual processing results of the similar screenshot frames are read for reuse; otherwise, the new screenshot frames are extracted to the high-risk exclusive task library for processing.

[0015] The above-mentioned method, device, equipment and medium for quickly capturing violations of live video broadcasts have the following technical effects compared with the prior art: 1. The machine review results output by the multi-dimensional machine review model are used to divide the violation risk levels of the live broadcast rooms. During manual review, the live broadcast rooms are sorted and displayed in order from high to low risk levels, giving full play to the advantages of the efficiency of machine review and the accuracy of manual review, and effectively ensuring the response time and efficiency of the live broadcast room violation capture and review.

[0016] 2. The live stream can be captured and recorded synchronously, and the real-time images not covered by the screenshot frames can be further reviewed for violations by replaying the recorded broadcast files, reducing the probability of misjudgment and missed judgment of violations in the live broadcast room.

[0017] 3. Prioritize review for high-risk live broadcast rooms and arrange dedicated reviewers to handle them, which improves the efficiency of illegal capture in high-risk live broadcast rooms. In addition, by storing the manual processing results of high-risk screenshot frames, when new screenshot frames are generated later, the manual processing results of previous similar high-risk screenshot frames are directly reused in the new screenshot frames. Reviewers do not need to remember the review operation information of repeated illegal screens, which reduces the workload of manual review and further improves the efficiency of violation review. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flowchart of a method for quickly capturing violations of live video broadcasting in one embodiment; Figure 2 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0020] In one embodiment, Figure 1 As shown, a method for quickly capturing violations of live video broadcast is provided, comprising the following steps: Step S1, obtain the live stream address after the user's live broadcast room starts broadcasting, and obtain the screenshot frames and recorded broadcast files during the user's live broadcast process by capturing and recording the live stream.

[0021] By synchronously capturing and recording the live stream, it is possible to further conduct illegal capture and review by replaying the recorded broadcast files in the real-time images not covered by the screenshot frames. This solves the problem that screenshot frames captured at a fixed frequency cannot fully cover the illegal content during the live broadcast, and reduces the probability of misjudgment or missed judgment of live broadcast violations.

[0022] Step S2, input the screenshot frame into the multi-dimensional machine review model for violation review, and divide the live broadcast room where the screenshot frame is located into high, medium and low risks according to the machine review results; wherein the machine review results include violation review marks and violation review confidence levels, and violation review marks include violation labels and no violation labels.

[0023] The use of a multi-dimensional machine review model can quickly identify violations in captured screenshot frames from multiple aspects such as pornography, politics, violence and terrorism, and vulgarity, thereby improving the efficiency of video live broadcast violation capture and review. In addition, the live broadcast rooms are divided into high, medium and low risks based on the machine review results, which facilitates the priority display of live broadcast rooms and screenshot frames with higher violation risks to reviewers during manual review, thereby effectively ensuring the response processing time of reviewers and improving the efficiency of high-risk violation reviews.

[0024] Step S3: In the review hall, the live broadcast rooms are sorted and displayed in order of risk level from high to low, and auditors are assigned to conduct manual review. For the medium and low risk live broadcast rooms, all the screenshot frames, real-time images and recorded broadcast files of the live broadcast room are provided to the auditors for review through the live broadcast details page, and the screenshot frames with violation labels are displayed first for review. If the manual review result is that there is a violation, the corresponding review process is carried out for the violation type of the screenshot frame; if the manual review result is that there is no violation, the violation label of the screenshot frame is removed, and the risk level of the live broadcast room is reduced to the same level as that of the room. The order in which the live broadcast rooms are displayed in the review hall; for high-risk live broadcast rooms, screenshot frames with violation labels and high review confidence are extracted to the high-risk exclusive task library, and exclusive reviewers are arranged to perform corresponding review processing according to the violation type, and the manual processing results of the screenshot frames are stored. When new screenshot frames with violation labels and high review confidence are generated in the same high-risk live broadcast room, by comparing the similarity between the new and old screenshot frames, if similar screenshot frames exist, the manual processing results of the similar screenshot frames are read for reuse; otherwise, the new screenshot frames are extracted to the high-risk exclusive task library for processing.

[0025] Specifically, in the review hall, live broadcast review tasks are evenly distributed according to the number of reviewers entering, and the review hall is refreshed in real time at a fixed frequency to ensure the latest status of the live broadcast room.

[0026] Through the combination of man and machine to conduct video live broadcast violation review, we can give full play to the advantages of the efficiency of machine review and the accuracy of manual review, and effectively ensure the response time and efficiency of live broadcast room violation capture review. In addition, by storing the manual processing results of high-risk screenshot frames, when new screenshot frames are generated later, the manual processing results of previous similar high-risk screenshot frames are directly reused to the new screenshot frames. Auditors do not need to remember the review operation information of repeated violation screens, which reduces the workload of manual review and further improves the efficiency of violation review.

[0027] The present application provides a method for quickly capturing violations in live video broadcasts. Through the combination of man and machine to conduct violation review, prioritize high-risk violations, reuse manual processing results of repeated violation images, and support playback and positioning viewing, it achieves efficient and accurate capture and processing of violations in live video broadcasts, and improves the security of live broadcast platform content.

[0028] In one embodiment, capturing and recording a live stream specifically includes the following steps: Get the user live stream address after the user's live broadcast room starts broadcasting, take screenshots of the live stream at a fixed frequency, save each frame of the image obtained as a screenshot frame, and number and name the screenshot frames in the form of self-increment of the number of screenshots to obtain continuous screenshot frames during the user's live broadcast. For example, the screenshot frames are named 1.jpg, 2.jpg, etc. If there are 100 pictures in total, there are f(n)=n+".jpg" screenshot frame files, where n is the number of screenshot frames.

[0029] The live stream is recorded into multiple slice files in ts (Transport Stream) format, and the slice files are incrementally written into the recorded files in m3u8 format in the order of the slice files' generation time, to obtain the recorded files during the user's live broadcast. Among them, the m3u8 file is a playlist file format, which is usually used to play audio and video content on the Internet. Each slice file in ts format usually contains only a few seconds of video content. The incremental writing of slice files can ensure that the recorded files in m3u8 format are always updated, which is suitable for scenarios with high real-time requirements such as live broadcast.

[0030] In one of the embodiments, the above method also includes: during the manual review process, for any screenshot frame suspected of violating the regulations in any live broadcast room, the time location of the screenshot frame is obtained based on the screenshot frame number and the fixed frequency of the screenshot frame, and the ts format slice file containing the screenshot frame time location is retrieved from the m3u8 format recording file of the same live broadcast room for playback and viewing, so as to further confirm whether the screenshot frame suspected of violating the regulations is in violation of the regulations.

[0031] In the above process, by calculating and obtaining the time location of the screenshot frame, the time when the screenshot frame was generated can be quickly obtained, and the ts format slice file containing the screenshot frame time location in the m3u8 format recorded broadcast file of the same live broadcast room can be retrieved for playback and viewing, thereby avoiding the need to scan the complete recorded broadcast file frame by frame. The ts format slice file usually only contains a few seconds of video content. By viewing the ts format slice file containing the screenshot frame time location, it can be seen whether there is any violation in the playback screen near the screenshot frame, thereby improving the efficiency of the screenshot frame playback positioning and improving the comprehensiveness and accuracy of the video live broadcast violation review.

[0032] In one embodiment, the screenshot frame is input into a multi-dimensional machine review model for violation review, and the live broadcast room where the screenshot frame is located is divided into high, medium and low risks according to the machine review results, including: The screenshot frame is input into the multi-dimensional machine review model for violation review, and the violation review mark and violation review confidence of the screenshot frame are output; among them, the multi-dimensional machine review model involves violation identification of pornographic, political, violent and vulgar content; For screenshot frames that hit no violation labels, the live broadcast room where the screenshot frame is located is judged as a low-risk live broadcast room; for screenshot frames that hit violation labels and the violation review confidence is within the low confidence interval, the live broadcast room where the screenshot frame is located is judged as a medium-risk live broadcast room; for screenshot frames that hit violation labels and the violation review confidence is within the high confidence interval, the live broadcast room where the screenshot frame is located is judged as a high-risk live broadcast room.

[0033] Furthermore, the calculation logic of the high and low confidence intervals confirmed by the multi-dimensional machine review model is: First, the screenshot frame data set to be reviewed is sent to the auditor and the multi-dimensional machine review model for violation review, and the manual review results are compared to determine whether they are consistent with the machine review results. If they are consistent, the machine review results are considered valid; otherwise, the machine review results are considered invalid. For example, if the manual review result is "licking the tongue", the corresponding correct classification is "vulgar". If the machine review result is also "vulgar", then the machine review result is considered valid, otherwise it is marked as invalid.

[0034] Then, all the machine review results of the screenshot frame data set to be reviewed output by the multi-dimensional machine review model are counted, and the intervals are grouped according to the violation review confidence scores of all machine review results. The total number of machine review results in each confidence interval is obtained, and the valid machine review results and invalid machine review results are distinguished.

[0035] Finally, the violation review efficiency of each confidence interval is obtained based on the number of valid machine review results and the total number of machine review results in each confidence interval. The confidence interval in which the violation review efficiency reaches the preset threshold is set as a high confidence interval; otherwise, the confidence interval in which the violation review efficiency is lower than the preset threshold is set as a low confidence interval.

[0036] In one embodiment, the manual processing results of the screenshot frames are stored. When a new screenshot frame with a violation label and a high audit confidence is generated in the same high-risk live broadcast room, by comparing the similarity between the new and old screenshot frames, if there are similar screenshot frames, the manual processing results of the similar screenshot frames are read for reuse; otherwise, the new screenshot frame is extracted to the high-risk exclusive task library for processing, including: The manual processing results and hash values ​​of the screenshot frames that hit the violation label and have high audit confidence are stored. When a new screenshot frame that hits the violation label and has high audit confidence is generated in the same high-risk live broadcast room, the hash value of the new screenshot frame is compared with the hash value of the screenshot frame stored in the high-risk processing result table to see if they are close. If the hash values ​​are close, read and reuse the manual processing results of the screenshot frames with close hash values ​​to the new screenshot frame for processing; otherwise, extract the new screenshot frame to the high-risk exclusive task library for review and processing; among them, the manual processing results include pass, warning, and blacklisting, etc.

[0037] By comparing whether the hash values ​​of the screenshot frames are close, the similarity of high-risk screenshot frames can be determined. For similar high-risk screenshot frames, the previous manual processing results can be directly reused, avoiding repeated review of similar illegal images, reducing the workload of manual review, and improving the efficiency of violation processing in high-risk live broadcast rooms.

[0038] In one embodiment, a device for quickly capturing violations of live video broadcast is provided, comprising: The frame capture and recording module is used to obtain the live stream address after the user's live broadcast room starts broadcasting, and obtain the screenshot frames and recorded broadcast files during the user's live broadcast by capturing and recording the live stream; The machine review module is used to input the screenshot frame into the multi-dimensional machine review model for violation review, and classify the live broadcast room where the screenshot frame is located into high, medium and low risks according to the machine review results; the machine review results include violation review marks and violation review confidence, and the violation review marks include violation labels and no violation labels; The manual review module is used to sort and display the live broadcast rooms in the review hall in order of risk level from high to low, and assign auditors to conduct manual review; among them, for medium and low-risk live broadcast rooms, all screenshot frames, real-time images and recorded broadcast files of the live broadcast room are provided to auditors for review through the live broadcast details page, and the screenshot frames with violation labels are displayed first for review. If the manual review result is that there is a violation, the corresponding review and processing will be carried out for the violation type of the screenshot frame; if the manual review result is that there is no violation, the violation label of the screenshot frame will be removed and the risk level of the live broadcast room will be reduced. and the order in which the live broadcast room is displayed in the review hall; for high-risk live broadcast rooms, extract the screenshot frames with violation labels and high review confidence to the high-risk exclusive task library, arrange exclusive reviewers to perform corresponding review processing according to the violation type, and store the manual processing results of the screenshot frames. When new screenshot frames with violation labels and high review confidence are generated in the same high-risk live broadcast room, by comparing the similarity between the new and old screenshot frames, if similar screenshot frames exist, read the manual processing results of the similar screenshot frames for reuse; otherwise, extract the new screenshot frames to the high-risk exclusive task library for processing.

[0039] Furthermore, the device also includes: a playback positioning module, which is used to calculate the time positioning of the screenshot frame based on the screenshot frame number and the fixed frequency of the screenshot frame for any screenshot frame suspected of violating the regulations in any live broadcast room during the manual review process, and to replay and view the ts format slice file containing the screenshot frame time positioning in the m3u8 format recording file of the same live broadcast room to further confirm whether the screenshot frame suspected of violating the regulations has violated the regulations.

[0040] For the specific definition of the device for quickly capturing violations of live video broadcast, please refer to the definition of the method for quickly capturing violations of live video broadcast above, which will not be repeated here. Each module in the above-mentioned device for quickly capturing violations of live video broadcast can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0041] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 2As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for quickly capturing violations of live video broadcasts is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0042] Those skilled in the art will understand that Figure 2 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0043] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented: Get the live stream address after the user's live broadcast room starts broadcasting, and obtain the screenshot frames and recorded broadcast files during the user's live broadcast by capturing and recording the live stream; The screenshot frames are input into the multi-dimensional machine review model for violation review, and the live broadcast room where the screenshot frames are located is divided into high, medium and low risks according to the machine review results; the machine review results include violation review marks and violation review confidence, and the violation review marks include violation labels and no violation labels; In the review hall, the live broadcast rooms are sorted and displayed in order of risk level from high to low, and auditors are assigned to conduct manual review; Among them, for medium- and low-risk live broadcast rooms, all screenshot frames, real-time images and recorded broadcast files of the live broadcast room are provided to the auditors for review through the live broadcast details page, and the screenshot frames with violation labels are displayed first for review. If the manual review result is that there is a violation, the corresponding review and processing will be carried out according to the violation type of the screenshot frame; if the manual review result is that there is no violation, the violation label of the screenshot frame will be removed, and the risk level of the live broadcast room and the display order of the live broadcast room in the review hall will be lowered; For high-risk live broadcast rooms, screenshot frames with violation labels and high audit confidence are extracted to the high-risk exclusive task library, and exclusive auditors are arranged to perform corresponding audit processing according to the violation type, and the manual processing results of the screenshot frames are stored. When new screenshot frames with violation labels and high audit confidence are generated in the same high-risk live broadcast room, the similarity between the new and old screenshot frames is compared. If similar screenshot frames exist, the manual processing results of the similar screenshot frames are read for reuse; otherwise, the new screenshot frames are extracted to the high-risk exclusive task library for processing.

[0044] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented: Get the live stream address after the user's live broadcast room starts broadcasting, and obtain the screenshot frames and recorded broadcast files during the user's live broadcast by capturing and recording the live stream; The screenshot frames are input into the multi-dimensional machine review model for violation review, and the live broadcast room where the screenshot frames are located is divided into high, medium and low risks according to the machine review results; the machine review results include violation review marks and violation review confidence, and the violation review marks include violation labels and no violation labels; In the review hall, the live broadcast rooms are sorted and displayed in order of risk level from high to low, and auditors are assigned to conduct manual review; Among them, for medium- and low-risk live broadcast rooms, all screenshot frames, real-time images and recorded broadcast files of the live broadcast room are provided to the auditors for review through the live broadcast details page, and the screenshot frames with violation labels are displayed first for review. If the manual review result is that there is a violation, the corresponding review and processing will be carried out according to the violation type of the screenshot frame; if the manual review result is that there is no violation, the violation label of the screenshot frame will be removed, and the risk level of the live broadcast room and the display order of the live broadcast room in the review hall will be lowered; For high-risk live broadcast rooms, screenshot frames with violation labels and high audit confidence are extracted to the high-risk exclusive task library, and exclusive auditors are arranged to perform corresponding audit processing according to the violation type, and the manual processing results of the screenshot frames are stored. When new screenshot frames with violation labels and high audit confidence are generated in the same high-risk live broadcast room, the similarity between the new and old screenshot frames is compared. If similar screenshot frames exist, the manual processing results of the similar screenshot frames are read for reuse; otherwise, the new screenshot frames are extracted to the high-risk exclusive task library for processing.

[0045] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0046] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0047] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for quickly capturing violations of live video broadcast, characterized in that: The method comprises: Get the live stream address after the user's live broadcast room starts broadcasting, and obtain the screenshot frames and recorded broadcast files during the user's live broadcast by capturing and recording the live stream; Input the screenshot frame into a multi-dimensional machine review model for violation review, and divide the live broadcast room where the screenshot frame is located into high, medium and low risks according to the machine review result; wherein the machine review result includes a violation review mark and a violation review confidence, and the violation review mark includes a violation label and a no violation label; In the review hall, the live broadcast rooms are sorted and displayed in order of risk level from high to low, and auditors are assigned to conduct manual review; Among them, for medium- and low-risk live broadcast rooms, all screenshot frames, real-time images and recorded broadcast files of the live broadcast room are provided to the auditors for review through the live broadcast details page, and the screenshot frames with violation labels are displayed first for review. If the manual review result is that there is a violation, the corresponding review and processing will be carried out according to the violation type of the screenshot frame; if the manual review result is that there is no violation, the violation label of the screenshot frame will be removed, and the risk level of the live broadcast room and the display order of the live broadcast room in the review hall will be lowered; For high-risk live broadcast rooms, screenshot frames with violation labels and high audit confidence are extracted to the high-risk exclusive task library, and exclusive auditors are arranged to perform corresponding audit processing according to the violation type, and the manual processing results of the screenshot frames are stored. When new screenshot frames with violation labels and high audit confidence are generated in the same high-risk live broadcast room, the similarity between the new and old screenshot frames is compared. If similar screenshot frames exist, the manual processing results of the similar screenshot frames are read for reuse; otherwise, the new screenshot frames are extracted to the high-risk exclusive task library for processing.

2. The method according to claim 1, characterized in that: Get the live stream address after the user's live broadcast room starts broadcasting, and obtain the screenshot frames and recorded broadcast files during the user's live broadcast by capturing and recording the live stream, including: Obtain the user live stream address after the user live broadcast room starts broadcasting, perform frame capture operation on the live stream at a fixed frequency, save each frame of the acquired image as a screenshot frame, and number and name the screenshot frames in the form of self-increment of the screenshot number, so as to obtain continuous screenshot frames during the user live broadcast process; The live stream is recorded as multiple slice files in ts format, and the slice files are incrementally written into the recorded broadcast file in m3u8 format according to the generation time sequence of the slice files to obtain the recorded broadcast file during the user's live broadcast process.

3. The method according to claim 2, characterized in that The method further comprises: During the manual review process, for any screenshot frame suspected of violating regulations in any live broadcast room, the time location of the screenshot frame is calculated based on the screenshot frame number and the fixed frequency of the screenshot frame, and the ts format slice file containing the screenshot frame time location in the m3u8 format recording file of the same live broadcast room is retrieved for playback and viewing to further confirm whether the screenshot frame suspected of violating regulations has violated regulations.

4. The method according to claim 1, characterized in that The screenshot frame is input into a multi-dimensional machine review model for violation review, and the live broadcast room where the screenshot frame is located is divided into high, medium and low risks according to the machine review results, including: Input the screenshot frame into a multi-dimensional machine review model for violation review, and output the violation review mark and violation review confidence of the screenshot frame; wherein the multi-dimensional machine review model involves violation identification of pornographic, political, violent, terrorist and vulgar content; For the screenshot frames that have no violation labels, the live broadcast room where the screenshot frames are located is determined to be a low-risk live broadcast room; For screenshot frames that hit violation labels and whose violation review confidence is in the low confidence interval, the live broadcast room where the screenshot frame is located is determined to be a medium-risk live broadcast room; For screenshot frames that hit the violation label and whose violation review confidence is within the high confidence interval, the live broadcast room where the screenshot frame is located is determined to be a high-risk live broadcast room.

5. The method according to claim 4, characterized in that The calculation logic of the multi-dimensional machine review model to confirm the high and low confidence intervals is: The screenshot frame data set to be reviewed is simultaneously sent to the auditor and the multi-dimensional machine review model for violation review, and a comparison is made to determine whether the manual review result is consistent with the machine review result. If they are consistent, the machine review result is determined to be valid; otherwise, the machine review result is determined to be invalid; Count all the machine review results of the screenshot frame data set to be reviewed output by the multi-dimensional machine review model, and group them into intervals according to the violation review confidence scores of all the machine review results, obtain the total number of machine review results in each confidence interval, and distinguish between valid machine review results and invalid machine review results; The violation review efficiency of each confidence interval is obtained by calculating the number of valid machine review results and the total number of machine review results in each confidence interval, and the confidence interval in which the violation review efficiency reaches the preset threshold is set as a high confidence interval; otherwise, the confidence interval in which the violation review efficiency is lower than the preset threshold is set as a low confidence interval.

6. The method according to claim 1, characterized in that The manual processing results of the screenshot frames are stored. When a new screenshot frame with a violation label and high audit confidence is generated in the same high-risk live broadcast room, the similarity between the new and old screenshot frames is compared. If similar screenshot frames exist, the manual processing results of the similar screenshot frames are read and reused; Otherwise, the new screenshot frame is extracted to the high-risk dedicated task library for processing, including: The manual processing results and hash values ​​of the screenshot frames that hit the violation label and have high audit confidence are stored. When a new screenshot frame that hits the violation label and has high audit confidence is generated in the same high-risk live broadcast room, the hash value of the new screenshot frame is compared with the hash value of the screenshot frame stored in the high-risk processing result table to see if they are close. If the hash values ​​are close, read and reuse the manual processing results of the screenshot frames with close hash values ​​to the new screenshot frame for processing; otherwise, extract the new screenshot frame to the high-risk exclusive task library for review and processing; wherein the manual processing results include pass, warning, and blacklist.

7. A device for quickly capturing violations of live video broadcast, characterized in that: The device comprises: The frame capture and recording module is used to obtain the live stream address after the user's live broadcast room starts broadcasting, and obtain the screenshot frames and recorded broadcast files during the user's live broadcast by capturing and recording the live stream; A machine review module, used to input the screenshot frame into a multi-dimensional machine review model for violation review, and divide the live broadcast room where the screenshot frame is located into high, medium and low risks according to the machine review result; wherein the machine review result includes a violation review mark and a violation review confidence, and the violation review mark includes a violation label and a no violation label; The manual review module is used to sort and display the live broadcast rooms in the review hall in order of risk level from high to low, and assign auditors to conduct manual review; among them, for medium and low-risk live broadcast rooms, all screenshot frames, real-time images and recorded broadcast files of the live broadcast room are provided to auditors for review through the live broadcast details page, and the screenshot frames with violation labels are displayed first for review. If the manual review result is that there is a violation, the corresponding review and processing will be carried out for the violation type of the screenshot frame; if the manual review result is that there is no violation, the violation label of the screenshot frame will be removed and the risk level of the live broadcast room will be reduced. and the order in which the live broadcast room is displayed in the review hall; for high-risk live broadcast rooms, extract the screenshot frames with violation labels and high review confidence to the high-risk exclusive task library, arrange exclusive reviewers to perform corresponding review processing according to the violation type, and store the manual processing results of the screenshot frames. When new screenshot frames with violation labels and high review confidence are generated in the same high-risk live broadcast room, by comparing the similarity between the new and old screenshot frames, if similar screenshot frames exist, read the manual processing results of the similar screenshot frames for reuse; otherwise, extract the new screenshot frames to the high-risk exclusive task library for processing.

8. The device according to claim 7, characterized in that The device also includes: a playback positioning module, which is used to calculate the time positioning of the screenshot frame according to the screenshot frame number and the fixed frequency of the screenshot frame for any screenshot frame suspected of violating the regulations in any live broadcast room during the manual review process, and to retrieve the ts format slice file containing the screenshot frame time positioning in the m3u8 format recording file of the same live broadcast room for playback and viewing, so as to further confirm whether the screenshot frame suspected of violating the regulations is in violation of the regulations.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images