Video content risk detection method and device, medium and equipment

Abstract

The embodiment of the invention discloses a video content risk detection method and device, a medium and equipment, and the method comprises the steps: firstly obtaining real-time video stream data, and extracting corresponding multi-modal detection data which comprises first information density data and second information density data of which the information density is greater than that of the first information density data; and then detecting the first information density data, determining a risk detection result, fusing the first information density data and the risk detection result into second information density data, and performing content decision on the fused second information density data to obtain a content risk decision result. According to the technical scheme, the multi-modal data is processed hierarchically through the information density, the detection efficiency of the risk content in the real-time video stream data can be effectively improved, and the risk detection timeliness is shortened; meanwhile, the repeated calculation amount of the high-information-density data in the content decision-making process can be reduced, and the content decision-making efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of video risk detection technology, and in particular to a method, device, medium and equipment for video content risk detection. Background Art

[0002] With the rapid development of live broadcasting and real-time video streaming services, the complexity and real-time requirements of live broadcasting and real-time video streaming have brought higher challenges to content review and risk control technology. Traditional content detection systems generally identify risks based on single-modal data, such as risk identification through frame capture images, audio slices, or keyword analysis. However, due to the diversity of information presentation and the complexity of risk factors in live broadcast content, it is difficult to accurately identify potential risk information by relying solely on single-modal detection, and often miss or misjudge due to insufficient information density.

[0003] At present, relevant technologies use multimodal data detection to improve risk identification capabilities. However, since this technical solution uses multimodal data for risk identification, the data complexity is relatively high, which may lead to duplicate data between multimodal data, resulting in a large number of repeated calculations and a waste of computing resources. At the same time, the time span from the occurrence of a risk to its detection is relatively long, and the risk detection efficiency is low. Summary of the invention

[0004] In a first aspect of the embodiments of the specification, a video content risk detection method is provided, which can reduce the data complexity of risk identification, reduce repeated calculations, save computing resources, shorten the time span from the occurrence of risk to detection, and improve the efficiency of risk detection. The method includes: Acquire real-time video stream data, and extract multimodal detection data corresponding to the real-time video stream data, wherein the multimodal detection data includes first information density data and second information density data, and the information density of the second information density data is greater than that of the first information density data; Testing the first information density data to determine a risk detection result; The first information density data and the risk detection result are fused into the second information density data, and content decision is performed on the fused second information density data to obtain a content risk decision result corresponding to the real-time video stream data.

[0005] Furthermore, in some embodiments, before fusing the first information density data and the risk detection result into the second information density data, the method also includes: inputting the risk detection results corresponding to each of the first information density data into the content decision system in order of output, to obtain preliminary risk decision results corresponding to each of the first information density data; if it is determined that the content risk level of any of the preliminary risk decision results is greater than or equal to the content risk level threshold, then the preliminary risk decision result is directly used as the content risk decision result corresponding to the real-time video stream data; if it is determined that the content risk level of each of the preliminary risk decision results is less than the content risk level threshold, then the first information density data and the risk detection result are fused into the second information density data.

[0006] Furthermore, in some embodiments, the content decision is performed on the fused second information density data to obtain a content risk decision result corresponding to the real-time video stream data, including: inputting the fused second information density data into a content decision system, so that the content decision system performs a content decision on the second information density data based on the risk detection result corresponding to the first information density data, to obtain an intermediate risk decision result of the fused second information density data; and determining the content risk decision result corresponding to the real-time video stream data according to each of the preliminary risk decision results and the intermediate risk decision results.

[0007] Further, in some embodiments, determining the content risk decision result corresponding to the real-time video stream data based on each of the preliminary risk decision results and the intermediate risk decision results includes: obtaining the multimodal risk detection weights corresponding to the first information density data and the second information density data; fusing the preliminary risk decision results and the intermediate risk decision results according to the multimodal risk detection weights to determine the content risk decision results corresponding to the real-time video stream data.

[0008] Further, in some embodiments, obtaining the multimodal risk detection weights corresponding to the first information density data and the second information density data includes: obtaining initial risk detection weights corresponding to the first information density data and the second information density data; updating the initial risk detection weights according to content risk decision results at historical moments associated with the real-time video stream data to obtain multimodal risk detection weights, so that the updated multimodal risk detection weights adapt to the video content of the real-time video stream data.

[0009] Furthermore, in some embodiments, the extracting of multimodal detection data corresponding to the real-time video stream data includes: obtaining a preset detection extraction time, the detection extraction time including a first extraction duration and a second extraction duration, the second extraction duration being greater than the first extraction duration; extracting the first information density data through the first extraction duration and in combination with the timestamp information of the real-time video stream data; extracting the second information density data through the second extraction duration and in combination with the timestamp information of the real-time video stream data.

[0010] Further, in some embodiments, the first extraction duration includes a video frame capture interval and an audio slice extraction duration; the first information density data is extracted through the first extraction duration and in combination with the timestamp information of the real-time video stream data, including: performing frame extraction processing on the real-time video stream data based on the video frame capture interval to obtain a frame capture image, and determining the timestamp information corresponding to the frame capture image according to the timestamp information of the real-time video stream data; performing audio extraction on the real-time video stream data based on the audio slice extraction duration to obtain audio slice data, and determining the timestamp information corresponding to the audio slice data according to the timestamp information of the real-time video stream data; performing speech recognition on the audio slice data to determine speech recognition text data, and determining the timestamp information corresponding to the speech recognition text data according to the timestamp information corresponding to the audio slice data; using the frame capture image, the audio slice data, the speech recognition text data and the corresponding timestamp information as the first information density data.

[0011] Further, in some embodiments, the second extraction duration includes a video slice extraction duration; the second information density data is extracted through the second extraction duration and in combination with the timestamp information of the real-time video stream data, including: performing video extraction on the real-time video stream data based on the video slice extraction duration to obtain video slice data, and determining the absolute start timestamp and absolute end timestamp corresponding to the video slice data according to the timestamp information of the real-time video stream data.

[0012] Further, in some embodiments, the first information density data includes at least a screenshot image, audio slice data and speech recognition text data; the detecting the first information density data to determine the risk detection result includes: performing feature analysis on the screenshot image to obtain screenshot image features, and determining the screenshot detection result through the screenshot image features; performing audio analysis on the audio slice data to obtain musical content features, and determining the audio detection result through the musical content features; performing content analysis on the speech recognition text data to obtain keyword features, and determining the text detection result through the keyword features; storing the screenshot detection result, the audio detection result and the text detection result as the risk detection result corresponding to the first information density data.

[0013] Further, in some embodiments, the fusion of the first information density data and the risk detection result into the second information density data includes: obtaining an absolute start timestamp and an absolute end timestamp corresponding to the second information density data; matching the first information density data within a corresponding timestamp range based on the absolute start timestamp and the absolute end timestamp; and integrating the matched first information density data and the corresponding risk detection result into detection information associated with the second information density data based on the timestamp information corresponding to the matched first information density data, to obtain the fused second information density data.

[0014] In a second aspect of the embodiments of the specification, a video content risk detection device is also proposed, including: A multimodal detection data extraction module, used to obtain real-time video stream data, and extract multimodal detection data corresponding to the real-time video stream data, wherein the multimodal detection data includes first information density data and second information density data, and the information density of the second information density data is greater than that of the first information density data; A risk detection module, used to detect the first information density data and determine a risk detection result; A content risk decision module is used to fuse the first information density data and the risk detection result into the second information density data, and perform content decision on the fused second information density data to obtain a content risk decision result corresponding to the real-time video stream data.

[0015] According to a third aspect of the embodiments of this specification, a computer program product is provided. The computer program product stores at least one instruction, and the at least one instruction is suitable for being loaded by a processor and executing the method steps in the first aspect.

[0016] According to a fourth aspect of the embodiments of this specification, a storage medium is provided, wherein the storage medium stores a computer program, and the computer program is suitable for being loaded by a processor and executing the steps of the method in the first aspect.

[0017] According to a fifth aspect of the embodiments of the present specification, there is also provided an electronic device, comprising: a processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the steps of the method in the first aspect.

[0018] In an embodiment of the present specification, multimodal detection data corresponding to real-time video stream data is extracted, and the multimodal detection data is divided into first information density data and second information density data according to information density, and the information density of the second information density data is greater than that of the first information density data; the first information density data is detected to determine a risk detection result, and the first information density data and the risk detection result are fused into the second information density data, and content decision is made on the fused second information density data to obtain a content risk decision result corresponding to the real-time video stream data. On the one hand, by processing the multimodal detection data in layers according to information density and performing prior risk decision processing on the first information density data, it is possible to quickly discover risk content in real-time video stream data, avoid the greater computational pressure of high information density data, reduce the computational complexity of the overall detection process, shorten the time span from risk occurrence to detection, and improve risk detection efficiency, thereby improving the response speed to risk content in real-time video stream data and ensuring network data security; on the other hand, after the risk detection result of the first information density data is obtained through prior risk decision processing, the risk detection result of the first information density data is fused with the second information density data. At this time, if there is repeated modal data in the second information density data, the risk detection result of the first information density data can be directly used as the risk detection result of the repeated data, avoiding a large amount of repeated calculations caused by risk decisions on repeated data, saving computing resources, and at the same time improving the risk decision efficiency of the second information density data, while ensuring the accuracy and stability of the risk detection results, further improving the risk detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 A schematic diagram of a system architecture of an exemplary application environment in which a video content risk detection method and apparatus according to an embodiment of the present specification can be applied is shown.

[0020] Figure 2 A flowchart of a video content risk detection method provided in an embodiment of this specification.

[0021] Figure 3 A flowchart of implementing content risk detection based on multimodal detection data is provided in an embodiment of this specification.

[0022] Figure 4 A schematic diagram of a process for dynamically updating and adjusting multimodal risk detection weights provided in an embodiment of this specification.

[0023] Figure 5 A schematic diagram of a process for determining a risk detection result of first information density data provided in an embodiment of this specification.

[0024] Figure 6 A schematic diagram of a process for fusing first information density data and risk detection results thereof into second information density data is provided in an embodiment of this specification.

[0025] Figure 7 A flowchart of a method for implementing risk detection by hierarchically processing multimodal detection data according to information density is provided in an embodiment of this specification.

[0026] Figure 8 A schematic diagram of the structure of a video content risk detection device provided in an embodiment of this specification.

[0027] Fig. 9 A schematic diagram of the structure of an electronic device provided in an embodiment of this specification. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solutions and advantages of this specification more clear, the technical solutions of this specification will be clearly and completely described below in combination with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this specification.

[0029] Figure 1 A schematic diagram of a system architecture of an exemplary application environment in which a method and device for detecting spoofed documents according to an embodiment of the present specification can be applied is shown.

[0030] like Figure 1 As shown, the system architecture 100 may include one or more of terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc. The terminal devices 101, 102, 103 may be various electronic devices with artificial intelligence (AI) computing capabilities, including but not limited to desktop computers, portable computers, smart phones, tablet computers, etc. It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. According to the implementation requirements, there may be any number of terminal devices, networks and servers. For example, the server 105 may be a server cluster composed of multiple servers.

[0031] The video content risk detection method provided in the embodiments of this specification is generally executed by the server 105, and accordingly, the video content risk detection device is generally set in the server 105. However, it is easy for those skilled in the art to understand that the video content risk detection method provided in the embodiments of this specification can also be executed by the terminal devices 101, 102, and 103, and accordingly, the video content risk detection device can also be set in the terminal devices 101, 102, and 103, which is not particularly limited in this exemplary embodiment.

[0032] See also Figure 2 , which is a flowchart of a video content risk detection method provided in the embodiment of this specification. In the embodiment of this specification, the video content risk detection method can be applied to a terminal device or a server, and this exemplary embodiment does not make any special restrictions on this. Figure 2 The process shown is described in detail. The video content risk detection method in the embodiment of this specification may specifically include the following steps: Step S210, acquiring real-time video stream data, and extracting multimodal detection data corresponding to the real-time video stream data, wherein the multimodal detection data includes first information density data and second information density data, and the information density of the second information density data is greater than that of the first information density data; Step S220, detecting the first information density data to determine a risk detection result; Step S230, integrating the first information density data and the risk detection result into the second information density data, and performing content decision on the integrated second information density data to obtain a content risk decision result corresponding to the real-time video stream data.

[0033] According to the video content risk detection method in the embodiment of this specification, on the one hand, by hierarchically processing the multimodal detection data through information density and performing prior risk decision processing on the first information density data, it is possible to quickly discover risk content in real-time video stream data, avoid the large calculation pressure of high information density data, reduce the calculation complexity in the overall detection process, shorten the time span from the occurrence of risk to its detection, and improve the risk detection efficiency, thereby improving the response speed to risk content in real-time video stream data and ensuring network data security; on the other hand, after the risk detection result of the first information density data is obtained through prior risk decision processing, the risk detection result of the first information density data is fused with the second information density data. At this time, if there is repeated modal data in the second information density data, the risk detection result of the first information density data can be directly used as the risk detection result of the repeated data, avoiding a large amount of repeated calculations caused by risk decisions on repeated data, saving computing resources, and at the same time improving the risk decision efficiency of the second information density data, while ensuring the accuracy and stability of the risk detection results, further improving the risk detection efficiency.

[0034] The video content risk detection method in the embodiments of this specification is described in detail below.

[0035] In step S210, real-time video stream data is acquired, and multimodal detection data corresponding to the real-time video stream data is extracted, wherein the multimodal detection data includes first information density data and second information density data, and the information density of the second information density data is greater than that of the first information density data.

[0036] In an example embodiment of the present specification, real-time video stream data refers to a continuous sequence of video frames transmitted through a network, which includes metadata such as video signals, audio signals, and timestamps. For example, real-time video stream data can be a live video stream or a real-time chat video in a chat tool. This example embodiment does not specifically limit the type of real-time video stream data.

[0037] Multimodal detection data refers to detection data with different modalities extracted from real-time video stream data. For example, multimodal detection data can be screenshot images, comment information, barrage information, or audio slice data, speech recognition text data, and video slice data. This example embodiment does not specifically limit the type of multimodal data that can be extracted from real-time video stream data.

[0038] Multimodal detection data can be extracted from real-time video stream data through a variety of related technologies, and this embodiment does not specifically limit this. For example, the real-time video stream data can be framed by a streaming media parser, and interval frame images can be extracted as cut-frame images; the audio stream can be extracted by an audio decoder and short-time audio slices can be generated to obtain audio slice data; the audio slice data can be converted into speech recognition text data by ASR (Automatic Speech Recognition) technology, and the video slice data can be extracted from the real-time video stream data by a video slicing tool. It can be understood that each extracted multimodal detection data can have its corresponding timestamp information at the same time, and the timestamp information can be used to mark the start and end time of each modal data to ensure the timing consistency of subsequent data matching.

[0039] In the embodiments of the present specification, the multimodal detection data can be divided into first information density data and second information density data according to information density, and the information density of the second information density data is greater than that of the first information density data. For example, the multimodal detection data can be data such as screenshot images, comment information, bullet screen information, audio slice data, voice recognition text data, video slice data, and the like, and the information density of the video slice data is greater than that of the screenshot images, comment information, bullet screen information, audio slice data, voice recognition text data, and the like. Therefore, the screenshot images, comment information, bullet screen information, audio slice data, voice recognition text data, and the like can be used as the first information density data, and the video slice data can be used as the second information density data. Of course, the data belonging to the first information density data and the second information density data can also be pre-divided according to the data repetition situation. For example, the multimodal detection data can include screenshot images, audio slice data, voice recognition text data, and video slice data, and the screenshot images are part of the video slice data, and the voice recognition text data are part of the audio slice data. Therefore, the screenshot images and the voice recognition text data can be used as the first information density data, and the audio slice data and the video slice data can be used as the second information density data. It is understandable that the division method and division content of the multimodal detection data of the first information density data and the second information density data can be customized according to the actual application scenario, and this embodiment does not specifically limit this. For the sake of convenience, the embodiments of this specification are described later by taking the first information density data as framed images, audio slice data, and speech recognition text data, and the second information density data as video slice data as an example.

[0040] In step S220, the first information density data is detected to determine a risk detection result.

[0041] In an example embodiment of the present specification, the risk detection result refers to the detection result obtained by performing risk content detection in combination with the content information contained in the first information density data. For example, assuming that the first information density data may include screenshot images, audio slice data, and speech recognition text data; specific risk object image features can be extracted through a deep learning model (such as a Faster R-CNN network or a YOLO network model, etc.) or a traditional image processing algorithm, and matched with a preset risk feature library to determine the risk detection result of the screenshot image; time domain and frequency domain features of audio slice data can be extracted through acoustic analysis technology (such as MFCC (Mel Frequency Cepstrum Coefficient) or pitch, timbre, tone features, etc.), and the extracted time domain and frequency domain features are compared and analyzed with sensitive audio templates to obtain the risk detection result of the audio slice data; natural language processing technology (such as keyword matching or semantic analysis) can be used to identify potential risk content in speech recognition text data. It is understandable that what is provided here is only one possible detection method for risk content detection of the first information density data. Other types of risk detection methods may also be used to detect the first information density data and determine the risk detection result. This example embodiment is not limited to this.

[0042] Risk content detection can be performed on each data in the first information density data in turn, and the specific detection order can be determined according to the extraction speed of each first information density data. For example, the screenshot image is extracted before the audio slice data and the speech recognition text data, and the risk detection is performed directly after the screenshot image is obtained. That is, the first information density data is only a classification of the multimodal detection data, and does not bind the various data in the first information density data into a whole. Each data in the first information density data is still detected independently.

[0043] It can be understood that in one embodiment, after obtaining the risk detection result of any first information density data, the first information density data and the corresponding risk detection result can be directly input into the content decision system for content decision making to determine the risk decision result of the current real-time video stream data, which can effectively improve the risk detection efficiency of the real-time video stream data and shorten the risk detection time limit, wherein the risk detection time limit is the time taken to detect potential risk information in the content, that is, the time span from the occurrence of the risk to its detection. In another embodiment, after waiting for each modal detection data contained in the first information density data to complete the risk detection, all the first information density data and the corresponding risk detection results can be input into the content decision system for content decision making to determine the risk decision result of the current real-time video stream data, which can effectively improve the accuracy and stability of the content decision result of the real-time video stream data.

[0044] When the content decision result of the first information density data is risk-free content, the first information density data and its corresponding risk detection result and timestamp information can be stored in a pre-set content database, waiting for subsequent call and processing. Optionally, the content database can be configured with a timestamp index structure to accelerate data matching, or the structure of the content database is a query database (such as an SQL database).

[0045] In step S230, the first information density data and the risk detection result are fused into the second information density data, and content decision is performed on the fused second information density data to obtain a content risk decision result corresponding to the real-time video stream data.

[0046] In an example embodiment of the present specification, the core of data fusion is to match the timestamp range of the first information density data with the second information density data, and associate the risk detection result of the first information density data with the second information density data. Accurate matching of timestamps is the key to data fusion, and modal association within the time series range can be achieved by establishing an index or querying a database. For example, the timestamp range of the second information density data, such as video slice data, can be used to retrieve the corresponding timestamps of the first information density data, such as frame images, audio slice data, and speech recognition text data, according to the timestamp range.

[0047] By processing multimodal detection data in layers according to information density and performing prior risk decision processing on the first information density data, it is possible to quickly discover risk content in real-time video stream data, avoid the greater computational pressure of high information density data, reduce the computational complexity of the overall detection process, shorten the time span from risk occurrence to detection, and improve risk detection efficiency, thereby improving the response speed to risk content in real-time video stream data and ensuring network data security. In addition, after the risk detection result of the first information density data is obtained through prior risk decision processing, the risk detection result of the first information density data is fused with the second information density data. At this time, if there is repeated modal data in the second information density data, the risk detection result of the first information density data can be directly used as the risk detection result of the repeated data, avoiding a large amount of repeated calculations caused by risk decisions on repeated data, saving computing resources, and at the same time improving the risk decision efficiency of the second information density data, while ensuring the accuracy and stability of the risk detection results, further improving the risk detection efficiency.

[0048] The contents of step S210 to step S230 are described below.

[0049] In an exemplary embodiment of the present specification, before the first information density data and the risk detection result are merged into the second information density data, content decision can be implemented according to the risk detection result of the first information density data, referring to Figure 3 As shown, this can be achieved through the following steps: Step S310, inputting the risk detection results corresponding to each of the first information density data into the content decision system in order of output, and obtaining the preliminary risk decision results corresponding to each of the first information density data; Step S320: If it is determined that the content risk level of any of the preliminary risk decision results is greater than or equal to the content risk level threshold, the preliminary risk decision result is directly used as the content risk decision result corresponding to the real-time video stream data; Step S330: If it is determined that the content risk level of each of the preliminary risk decision results is less than the content risk level threshold, the first information density data and the risk detection result are merged into the second information density data.

[0050] Among them, the output order refers to the order in which each first information density data is extracted from the real-time video stream data. For example, the first information density data may include screenshot images, audio slice data and speech recognition text data, wherein the screenshot images may be video frame images extracted from the real-time video stream data at intervals, and the audio slice data is continuous audio data within a certain period of time in the real-time video stream data, and the speech recognition text data is the data obtained after recognizing the audio slice data. Therefore, there is a time difference in the output order of each first information density data.

[0051] The risk detection results corresponding to each first information density data and each first information density data can be input into the content decision system in sequence according to the output order of each first information density data to make risk content decisions, and obtain preliminary risk decision results corresponding to each first information density data. For example, the preliminary risk decision results can be no risk, low risk, medium risk and high risk, etc. Of course, the preliminary risk decision results can also be divided into levels 1 to 5, and the content risk level corresponding to level 1 is lower, and the content risk level corresponding to level 5 is higher. This example embodiment does not make any special limitations on the division method of the preliminary risk decision results.

[0052] When it is determined that the content risk level of any preliminary risk decision result is greater than or equal to the content risk level threshold, for example, the content risk level threshold may be medium risk or level 3, it can be considered that medium- and high-risk content appears in the real-time video stream data. The preliminary risk decision result can be directly used as the content risk decision result corresponding to the real-time video stream data, thereby achieving rapid discovery of medium- and high-risk content through the first information density data, effectively shortening the time span from risk occurrence to detection, and improving risk detection efficiency.

[0053] When it is determined that the content risk levels of each preliminary risk decision result are all less than the content risk level threshold, it can be considered that the first information density data has preliminarily detected the risk content and further confirmation is needed. The first information density data and the corresponding risk detection result can be fused to the second information density data, so as to achieve further detection of the risk content of the real-time video stream data. Since the information density of the second information density data is relatively high, the accuracy and stability of the risk content decision results can be guaranteed. At the same time, the second information density data is fused with the first information density data and the corresponding risk detection result. Risk content decisions can be made on the second information density data, avoiding detection and processing of repeated data, effectively reducing the amount of repeated calculations, saving computing resources, and further improving the efficiency of risk detection.

[0054] Optionally, after determining that the content risk level of each preliminary risk decision result is less than the content risk level threshold, and fusing the first information density data and the corresponding risk detection result to the second information density data, a content decision can be made on the fused second information density data to obtain a content risk decision result corresponding to the real-time video stream data. This can be achieved by the following steps: The fused second information density data can be input into the content decision system, so that the content decision system makes a content decision on the second information density data based on the risk detection result corresponding to the first information density data, and obtains an intermediate risk decision result of the fused second information density data, and then determines the content risk decision result corresponding to the real-time video stream data according to each preliminary risk decision result and the intermediate risk decision result. For example, the content risk decision result corresponding to the real-time video stream data can be obtained by weighted calculation based on the weights corresponding to each preliminary risk decision result and the intermediate risk decision result; of course, each preliminary risk decision result and the intermediate risk decision result can also be input into a pre-set standardized risk scoring function to comprehensively determine the content risk decision result corresponding to the real-time video stream data. This example embodiment does not specifically limit the specific method of determining the content risk decision result corresponding to the real-time video stream data according to each preliminary risk decision result and the intermediate risk decision result.

[0055] Optionally, the following steps may be used to determine the content risk decision result corresponding to the real-time video stream data according to each preliminary risk decision result and the intermediate risk decision result, including: The multimodal risk detection weights corresponding to the first information density data and the second information density data can be obtained, and the preliminary risk decision results and the intermediate risk decision results are fused according to the multimodal risk detection weights to determine the content risk decision results corresponding to the real-time video stream data.

[0056] For example, assuming that the first information density data may include frame images, audio segment data, and speech recognition text data, and the second information density data may include video segment data, the following relationship may be used to determine the content risk decision result corresponding to the real-time video stream data based on the preliminary risk decision results and the intermediate risk decision results: in, It can represent the content risk decision results, , , and The multimodal risk detection weights corresponding to the frame image, audio segment data, speech recognition text data and video segment data may be represented respectively. For example, the multimodal risk detection weights may be 0.2 for the frame image, 0.2 for the audio segment data, 0.2 for the speech recognition text data and 0.4 for the video segment data. It is to be understood that this is only a schematic example and may be customized according to the actual situation. , , The preliminary risk decision results corresponding to the first information density data can be represented respectively. The intermediate risk decision result of the second information density data can be represented.

[0057] By integrating the preliminary risk decision results and the intermediate risk decision results through multimodal risk detection weights, the content risk decision results corresponding to the real-time video stream data are obtained, which can effectively improve the accuracy and stability of the content risk decision results, realize the effective detection of real-time video stream data with risky content, and ensure the security of network data.

[0058] Optionally, you can Figure 4 The steps in the embodiment are used to update the multimodal risk detection weights corresponding to the first information density data and the second information density data, referring to Figure 4 As shown, it may specifically include: Step S410, obtaining initial risk detection weights corresponding to the first information density data and the second information density data; Step S420, based on the content risk decision results of the historical moments associated with the real-time video stream data, the initial risk detection weight is updated to obtain a multimodal risk detection weight, so that the updated multimodal risk detection weight is adapted to the video content of the real-time video stream data.

[0059] Among them, the initial risk detection weight refers to the multimodal risk detection weight corresponding to the first information density data and the second information density data which are pre-set based on experience. For example, the initial risk detection weight can be 0.2 for the screenshot image, 0.2 for the audio segment data, 0.2 for the speech recognition text data, and 0.4 for the video segment data. Of course, the initial risk detection weight can be customized according to the actual situation, and this embodiment is not limited to this.

[0060] After at least one round of content decision making for the first information density data and the second information density data, the content risk decision results of the historical moments associated with the real-time video stream data can be obtained. At this time, the content risk decision results can show the decisive factors in the first information density data and the second information density data in the historical moments. For example, if the content risk decision results in the historical moments are determined by the audio slice data in the first information density data, then it can be considered that the high-risk content in the current real-time video stream data is in the audio information, so the multimodal risk detection weight corresponding to the audio slice data can be adjusted, such as the multimodal risk detection weight can be adjusted to 0.1 for the screenshot image, 0.4 for the audio slice data, 0.2 for the speech recognition text data, and 0.3 for the video slice data; of course, other first information density data and second information density data are updated similarly, and the specific update ratio or update weight can be customized, and this embodiment does not make any special limitation on this.

[0061] After multiple rounds of content decision-making, the dynamically updated and adjusted multimodal risk detection weights will become increasingly adapted to the video content of the real-time video stream data, so that the content risk decision results determined based on the multimodal risk detection weights are more in line with the video content in the real-time video stream data, effectively improving the accuracy and effectiveness of the content risk decision results, ensuring the accuracy of risk processing of real-time video stream data, and avoiding false detection rates.

[0062] In an exemplary embodiment of the present specification, the following steps may be used to extract multimodal detection data corresponding to real-time video data, which may specifically include the following steps: A preset detection extraction time can be obtained, and the detection extraction time may include a first extraction duration and a second extraction duration, and the second extraction duration is greater than the first extraction duration; the first information density data is extracted through the first extraction duration and combined with the timestamp information of the real-time video stream data; the second information density data is extracted through the second extraction duration and combined with the timestamp information of the real-time video stream data.

[0063] Among them, the first extraction duration refers to the pre-set time interval data for extracting the first information density data in the real-time video stream data, and its duration is usually short to meet the real-time requirements. For example, the first extraction duration can be a video frame interval for extracting a framed image, or it can be an audio slice extraction duration for extracting audio slice data. The specific interval or duration can be customized. This embodiment does not impose any special limitation on the value setting of the first extraction duration.

[0064] The second extraction duration refers to the pre-set time interval data for extracting the second information density data in the real-time video stream data. The duration is longer to capture more complete video content. For example, the second extraction duration can be the video slice extraction duration of the video slice data, or it can be the duration between video segment nodes. The specific duration can be customized or set according to the application situation. This example embodiment does not impose any special limitation on the value setting of the second extraction duration.

[0065] The second extraction duration can be greater than the first extraction duration. For example, the initial values ​​of the first extraction duration and the second extraction duration can be defined through a system configuration file. For example, the video frame interval of the framed image can be set to 0.5 seconds, the audio slice extraction duration of the audio slice data can be set to 1 second, and the video slice extraction duration of the video slice data can be set to 3 seconds. Then, the start time and end time of each modal data can be marked with timestamp information to ensure data timing consistency.

[0066] It can be understood that the first extraction time and the second extraction time may not be fixed, and a real-time dynamic adjustment mechanism may be used to dynamically adjust the first extraction time and the second extraction time according to the current network delay and processing performance; of course, a machine learning model may be used to predict risk hotspot areas in real-time video stream data, and the sampling density of the risk hotspot areas may be increased, thereby dynamically adjusting the first extraction time and the second extraction time according to the sampling density; this embodiment does not specifically limit the specific real-time dynamic adjustment mechanism.

[0067] Optionally, the extraction of the first information density data and the second information density data may be achieved through the following steps, which may specifically include the following steps: Frame extraction processing can be performed on real-time video stream data based on the video frame capture interval to obtain a frame capture image, and the timestamp information corresponding to the frame capture image can be determined based on the timestamp information of the real-time video stream data. For example, after the real-time video stream data is decoded by a streaming media decoder, frame sampling can be performed on the decoded real-time video stream data in combination with the video frame capture interval to obtain a frame capture image, and the timestamp information in the real-time video stream data at the current sampling moment is used as the timestamp information corresponding to the frame capture image, and the frame capture image is time-sequentially marked in combination with the timestamp information to ensure the consistency of subsequent matching.

[0068] Audio extraction can be performed on real-time video stream data based on the audio slice extraction duration to obtain audio slice data, and the timestamp information corresponding to the audio slice data can be determined based on the timestamp information of the real-time video stream data. For example, the original audio data in the real-time video stream data can be extracted in real time by an audio decoder, and the original audio data can be audio segmented in combination with the audio slice extraction duration to obtain audio slice data, and the absolute start timestamp and absolute end timestamp corresponding to the audio slice data can be determined in combination with the timestamp information in the real-time video stream data at the current sampling moment and the audio slice extraction duration to ensure the consistency of subsequent matching and fusion.

[0069] Speech recognition can be performed on the audio slice data obtained by segmentation based on the audio slice extraction time length, and speech recognition text data can be determined. According to the timestamp information corresponding to the audio slice data, such as the absolute start timestamp and the absolute end timestamp, the timestamp information corresponding to the speech recognition text data can be determined, that is, the absolute start timestamp and the absolute end timestamp of the speech recognition text data are consistent with the audio slice data. Then, the frame image, the audio slice data, the speech recognition text data and the corresponding timestamp information can be used as the first information density data.

[0070] Video extraction can be performed on real-time video stream data based on the video slice extraction duration to obtain video slice data, and the absolute start timestamp and absolute end timestamp corresponding to the video slice data can be determined according to the timestamp information of the real-time video stream data.

[0071] By detecting the extraction time, the real-time video stream data with higher complexity can be divided into multimodal detection data, and the first extraction time and the second extraction time can be combined to realize the extraction of the first information density data and the second information density data, and the time stamp information corresponding to the real-time video stream data can be combined to realize the marking of the time series first information density data and the second information density data, so as to provide a time series basis for the subsequent fusion of multimodal detection data, improve the accuracy and efficiency of data fusion, and improve the efficiency of risk detection and the accuracy of risk content decision-making results.

[0072] In an exemplary embodiment of this specification, it is possible to Figure 5The steps in the above are used to detect the first information density data and determine the risk detection result. Figure 5 As shown, it may specifically include: Step S510, performing feature analysis on the frame-shot image to obtain frame-shot image features, and determining a frame-shot detection result according to the frame-shot image features; Step S520, performing audio analysis on the audio slice data to obtain musical content features, and determining an audio detection result according to the musical content features; Step S530, performing content analysis on the speech recognition text data to obtain keyword features, and determining text detection results based on the keyword features; Step S540: storing the frame detection result, the audio detection result and the text detection result as risk detection results corresponding to the first information density data.

[0073] Among them, the risk content features in the screenshot image can be extracted by an image feature extraction model based on a convolutional neural network to obtain the screenshot image features; the risk content features in the screenshot image can also be extracted based on traditional image processing technologies such as object recognition and edge segmentation to obtain the screenshot image features. This example embodiment does not specifically limit the feature analysis method of the screenshot image. After obtaining the screenshot image features, the screenshot image features can be compared with the risk image features in a preset risk content database to determine the screenshot detection result.

[0074] The acoustic feature extraction model based on the Transformer network can be used to extract and classify the musical features such as timbre and pitch contained in the audio slice data to obtain musical content features; of course, the Mel-frequency cepstral coefficients can also be used to extract and classify the musical features such as spectrum features and high-frequency components contained in the audio slice data to obtain musical content features. This example embodiment does not specifically limit the audio analysis method for audio slice data. After obtaining the musical content features, the musical content features can be compared with the risk musical content features in the preset risk content database to determine the audio detection results.

[0075] The content analysis of speech recognition text data can be achieved through a sentiment analysis model based on LSTM (Long Short Term Memory) and combined with a keyword extraction model to obtain keyword features, and the keyword features can be compared with the risk keyword features in a preset risk content database to determine the text detection results.

[0076] Through comprehensive detection of the first information density data, due to the different output orders of each data in the first information density data, it is possible to quickly detect risk content in real-time video stream data, effectively improve risk detection efficiency, shorten risk detection time, and when the detected risk level is low, it is integrated into the second information density data to assist in risk detection and content decision-making of the second information density data, effectively avoid the calculation of repeated data, reduce the amount of repeated calculations, save computing resources, and further improve risk detection efficiency while ensuring the accuracy of risk content decision-making results.

[0077] In an exemplary embodiment of this specification, it is possible to Figure 6 The steps in are used to integrate the first information density data and the risk detection results into the second information density data, refer to Figure 6 As shown, it may specifically include: Step S610, obtaining an absolute start timestamp and an absolute end timestamp corresponding to the second information density data; Step S620, matching first information density data within a corresponding timestamp range based on the absolute start timestamp and the absolute end timestamp; Step S630, according to the timestamp information corresponding to the matched first information density data, the matched first information density data and the corresponding risk detection result are integrated into the detection information associated with the second information density data to obtain the fused second information density data.

[0078] Among them, the absolute start timestamp refers to a timestamp used to mark the starting time point in the video slice data. The timestamp is based on the playback timeline of the real-time video stream data and is consistent with the overall time sequence of the real-time video stream data, ensuring that different modal data can be synchronized and associated through timestamps.

[0079] The absolute end timestamp refers to a timestamp used to mark the end time point of the video slice data. This timestamp marks the end time point of the video slice and is also based on the playback timeline of the real-time video stream to ensure that the end points of data in different modalities are consistent.

[0080] The absolute start timestamp and the absolute end timestamp mark the start time point and the end time point of the video slice data to keep them absolutely consistent with the playback time axis of the real-time video stream, which is different from the timestamp mark based on the system clock or the frame number.

[0081] By matching the absolute start timestamp and the absolute end timestamp to the corresponding timestamp range, matching the first information density data and the corresponding risk detection result to realize the association of modal data, it can be ensured that the second information density data has a repeated data portion that is completely aligned with the first information density data, so that the first information density data and its detection result can be directly used as the association information of the second information density data having the same part of data, thereby avoiding repeated calculation of this part of repeated data, effectively reducing the amount of repeated calculations, saving computing resources, and thus improving the efficiency of risk detection.

[0082] Figure 7 A flowchart of a method for implementing risk detection by hierarchically processing multimodal detection data according to information density is provided in an embodiment of this specification.

[0083] refer to Figure 7As shown, in step S710, real-time video stream data transmitted in real time is obtained; in step S720, the real-time video stream data can be processed by a preprocessing module to obtain multimodal detection data corresponding to the real-time video stream data; for example, frame extraction processing can be performed on the real-time video stream data based on the video frame capture interval to obtain a frame capture image, and the timestamp information corresponding to the frame capture image can be determined according to the timestamp information of the real-time video stream data; audio extraction can be performed on the real-time video stream data based on the audio slice extraction duration to obtain audio slice data, and the timestamp information corresponding to the audio slice data can be determined according to the timestamp information of the real-time video stream data; speech recognition can be performed on the audio slice data to determine speech recognition text data, and the timestamp information corresponding to the audio slice data can be determined according to the timestamp information of the real-time video stream data. The timestamp information corresponding to the audio slice data is used to determine the timestamp information corresponding to the speech recognition text data; the frame image, the audio slice data, the speech recognition text data and the corresponding timestamp information are used as the first information density data; video extraction can be performed on the real-time video stream data based on the video slice extraction duration to obtain video slice data, and the absolute start timestamp and the absolute end timestamp corresponding to the video slice data are determined according to the timestamp information of the real-time video stream data, and the video slice data and the corresponding absolute start timestamp and the absolute end timestamp are used as the second information density data; step S730, the risk detection results corresponding to each first information density data are sequentially input into the risk detection system in the output order for risk detection The preliminary risk decision results corresponding to each first information density data are measured and obtained, and the preliminary risk decision results obtained are input into the content decision system for content decision; for example, feature analysis can be performed on the screenshot image to obtain the screenshot image features, and the screenshot detection results can be determined by the screenshot image features; audio analysis can be performed on the audio slice data to obtain the rhythm content features, and the audio detection results can be determined by the rhythm content features; content analysis can be performed on the speech recognition text data to obtain the keyword features, and the text detection results can be determined by the keyword features; step S740, the screenshot detection results, the audio detection results and the text detection results are stored as the risk detection results corresponding to the first information density data, and stored in the database, and as The video slice data of the second information density data is stored in the database; if it is determined that the content risk level of any preliminary risk decision result is greater than or equal to the content risk level threshold, the preliminary risk decision result is directly used as the content risk decision result corresponding to the real-time video stream data, and the current process is terminated; if it is determined that the content risk level of each preliminary risk decision result is less than the content risk level threshold, the first information density data and the risk detection result are merged into the second information density data, and step S750 is executed; in step S750, the absolute start timestamp and the absolute end timestamp corresponding to the second information density data can be obtained; the first information density data within the corresponding timestamp range is matched based on the absolute start timestamp and the absolute end timestamp;According to the timestamp information corresponding to the matched first information density data, the matched first information density data and the corresponding risk detection result are integrated into the detection information associated with the second information density data to obtain the fused second information density data; step S760, the fused second information density data can be input into the content decision system, so that the content decision system makes a content decision on the second information density data based on the risk detection result corresponding to the first information density data, and obtains the intermediate risk decision result of the fused second information density data; according to each preliminary risk decision result and the intermediate risk decision result, the content risk decision result corresponding to the real-time video stream data is determined, and the current process ends. ;

[0084] To summarize, the video content risk detection method provided in the embodiments of the present specification extracts multimodal detection data corresponding to real-time video stream data, and divides the multimodal detection data into first information density data and second information density data according to information density, and the information density of the second information density data is greater than that of the first information density data; the first information density data is detected to determine the risk detection result, and the first information density data and the risk detection result are merged into the second information density data, and content decision is made on the merged second information density data to obtain a content risk decision result corresponding to the real-time video stream data. On the one hand, by processing the multimodal detection data in layers according to information density and performing prior risk decision processing on the first information density data, it is possible to quickly discover risk content in real-time video stream data, avoid the greater computational pressure of high information density data, reduce the computational complexity of the overall detection process, shorten the time span from risk occurrence to detection, and improve risk detection efficiency, thereby improving the response speed to risk content in real-time video stream data and ensuring network data security; on the other hand, after the risk detection result of the first information density data is obtained through prior risk decision processing, the risk detection result of the first information density data is fused with the second information density data. At this time, if there is repeated modal data in the second information density data, the risk detection result of the first information density data can be directly used as the risk detection result of the repeated data, avoiding a large amount of repeated calculations caused by risk decisions on repeated data, saving computing resources, and at the same time improving the risk decision efficiency of the second information density data, while ensuring the accuracy and stability of the risk detection results, further improving the risk detection efficiency.

[0085] See also Figure 8 , is a schematic diagram of the structure of a video content risk detection device provided in an embodiment of this specification. Figure 8As shown, the video content risk detection device 800 can be implemented as all or part of an electronic device through software, hardware, or a combination of both. According to some embodiments, the video content risk detection device 800 includes a multimodal detection data extraction module 810, a risk detection module 820, and a content risk decision module 830, specifically including: The multimodal detection data extraction module 810 may be used to obtain real-time video stream data and extract multimodal detection data corresponding to the real-time video stream data, wherein the multimodal detection data includes first information density data and second information density data, and the information density of the second information density data is greater than that of the first information density data; The risk detection module 820 may be used to detect the first information density data and determine a risk detection result; The content risk decision module 830 can be used to fuse the first information density data and the risk detection result into the second information density data, and perform content decision on the fused second information density data to obtain a content risk decision result corresponding to the real-time video stream data.

[0086] Optionally, the video content risk detection device 800 further includes a first information density data sequence processing module, and the first information density data sequence processing module is configured as follows: Inputting the risk detection results corresponding to each of the first information density data into the content decision system in order of output, and obtaining the preliminary risk decision results corresponding to each of the first information density data; If it is determined that the content risk level of any of the preliminary risk decision results is greater than or equal to the content risk level threshold, the preliminary risk decision result is directly used as the content risk decision result corresponding to the real-time video stream data; If it is determined that the content risk level of each of the preliminary risk decision results is less than the content risk level threshold, the first information density data and the risk detection result are merged into the second information density data.

[0087] Optionally, the content risk decision module 830 is configured to: Inputting the fused second information density data into the content decision system, so that the content decision system makes a content decision on the second information density data based on the risk detection result corresponding to the first information density data, and obtains an intermediate risk decision result of the fused second information density data; According to each of the preliminary risk decision results and the intermediate risk decision results, a content risk decision result corresponding to the real-time video stream data is determined.

[0088] Optionally, the content risk decision module 830 is configured to: Obtaining multimodal risk detection weights corresponding to the first information density data and the second information density data; The preliminary risk decision result and the intermediate risk decision result are integrated according to the multimodal risk detection weight to determine the content risk decision result corresponding to the real-time video stream data.

[0089] Optionally, the content risk decision module 830 is configured to: Obtaining initial risk detection weights corresponding to the first information density data and the second information density data; According to the content risk decision results of the historical moments associated with the real-time video stream data, the initial risk detection weight is updated to obtain a multimodal risk detection weight, so that the updated multimodal risk detection weight is adapted to the video content of the real-time video stream data.

[0090] Optionally, the multimodal detection data extraction module 810 is configured to: Obtaining a preset detection and extraction time, wherein the detection and extraction time includes a first extraction time length and a second extraction time length, wherein the second extraction time length is greater than the first extraction time length; Extracting the first information density data by combining the first extraction duration with the timestamp information of the real-time video stream data; The second information density data is extracted by combining the second extraction duration with the timestamp information of the real-time video stream data.

[0091] Optionally, the first extraction duration includes a video frame capture interval and an audio slice extraction duration; the multimodal detection data extraction module 810 is configured as follows: Performing frame extraction processing on the real-time video stream data based on the video frame capture interval to obtain a frame capture image, and determining timestamp information corresponding to the frame capture image according to timestamp information of the real-time video stream data; Performing audio extraction on the real-time video stream data based on the audio slice extraction duration to obtain audio slice data, and determining timestamp information corresponding to the audio slice data according to timestamp information of the real-time video stream data; Performing speech recognition on the audio slice data to determine speech recognition text data, and determining timestamp information corresponding to the speech recognition text data according to the timestamp information corresponding to the audio slice data; The frame-cut image, the audio slice data, the speech recognition text data and the corresponding timestamp information are used as the first information density data.

[0092] Optionally, the second extraction duration includes a video slice extraction duration; and the multimodal detection data extraction module 810 is configured to: The real-time video stream data is subjected to video extraction based on the video slice extraction duration to obtain video slice data, and an absolute start timestamp and an absolute end timestamp corresponding to the video slice data are determined according to timestamp information of the real-time video stream data.

[0093] Optionally, the first information density data includes at least a frame image, audio slice data and speech recognition text data; the risk detection module 820 is configured to: Performing feature analysis on the frame-shot image to obtain frame-shot image features, and determining a frame-shot detection result based on the frame-shot image features; Performing audio analysis on the audio slice data to obtain musical content features, and determining an audio detection result based on the musical content features; Performing content analysis on the speech recognition text data to obtain keyword features, and determining text detection results based on the keyword features; The frame detection result, the audio detection result and the text detection result are stored as risk detection results corresponding to the first information density data.

[0094] Optionally, the content risk decision module 830 is configured to: Obtaining an absolute start timestamp and an absolute end timestamp corresponding to the second information density data; Matching first information density data within a corresponding timestamp range based on the absolute start timestamp and the absolute end timestamp; According to the timestamp information corresponding to the matched first information density data, the matched first information density data and the corresponding risk detection result are integrated into the detection information associated with the second information density data to obtain the fused second information density data.

[0095] The above device embodiments correspond to the method embodiments. For specific descriptions, please refer to the description of the method embodiments, which will not be repeated here. The device embodiments are obtained based on the corresponding method embodiments and have the same technical effects as the corresponding method embodiments. For specific descriptions, please refer to the corresponding method embodiments.

[0096] The embodiments of this specification also provide a computer storage medium, which can store multiple instructions, and the instructions are suitable for being loaded by a processor and executing the method as described in the embodiments of this specification. The specific execution process can be found in the specific description of the embodiments of this specification, and will not be repeated here.

[0097] This specification also provides a computer program product, which stores at least one instruction, and the at least one instruction is loaded by the processor and executes the method as described in the embodiments of this specification. The specific execution process can be found in the specific description of the embodiments of this specification, and will not be repeated here.

[0098] The embodiments of this specification also provide Fig. 9 The structural diagram of the electronic device shown in FIG. Fig. 9 At the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the above-mentioned voice activity detection method.

[0099] Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is to say, the executor of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0100] For the improvement of a technology, it can be clearly distinguished whether it is a hardware improvement (for example, improvement of the circuit structure of diodes, transistors, switches, etc.) or a software improvement (improvement of the method flow). However, with the development of technology, many improvements of the method flow today can be regarded as direct improvements of the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that the improvement of a method flow cannot be implemented with a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming themselves, without having to ask chip manufacturers to design and make dedicated integrated circuit chips. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one kind of HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0101] The controller may be implemented in any suitable manner, for example, the controller may take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) executable by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller may also be implemented as part of the control logic of the memory. It is also known to those skilled in the art that, in addition to implementing the controller in a purely computer-readable program code manner, the controller may be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, such a controller may be considered as a hardware component, and the devices for implementing various functions included therein may also be considered as structures within the hardware component. Or even, the devices for implementing various functions may be considered as both software modules for implementing the method and structures within the hardware component.

[0102] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0103] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0104] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0105] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0106] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0107] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0108] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0109] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0110] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0111] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0112] It should be understood by those skilled in the art that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0113] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0114] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0115] The above description is only an embodiment of the present specification and is not intended to limit the present specification. For those skilled in the art, the present specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification shall be included in the scope of the claims of the present specification.

Claims

1. A video content risk detection method, the method comprising: Acquire real-time video stream data, and extract multimodal detection data corresponding to the real-time video stream data, wherein the multimodal detection data includes first information density data and second information density data, and the information density of the second information density data is greater than that of the first information density data; Testing the first information density data to determine a risk detection result; The first information density data and the risk detection result are fused into the second information density data, and content decision is performed on the fused second information density data to obtain a content risk decision result corresponding to the real-time video stream data.

2. The video content risk detection method according to claim 1, before fusing the first information density data and the risk detection result into the second information density data, the method further comprises: Inputting the risk detection results corresponding to each of the first information density data into the content decision system in order of output, and obtaining the preliminary risk decision results corresponding to each of the first information density data; If it is determined that the content risk level of any of the preliminary risk decision results is greater than or equal to the content risk level threshold, the preliminary risk decision result is directly used as the content risk decision result corresponding to the real-time video stream data; If it is determined that the content risk level of each of the preliminary risk decision results is less than the content risk level threshold, the first information density data and the risk detection result are merged into the second information density data.

3. According to the video content risk detection method of claim 2, the step of performing content decision on the fused second information density data to obtain a content risk decision result corresponding to the real-time video stream data comprises: Inputting the fused second information density data into the content decision system, so that the content decision system makes a content decision on the second information density data based on the risk detection result corresponding to the first information density data, and obtains an intermediate risk decision result of the fused second information density data; According to each of the preliminary risk decision results and the intermediate risk decision results, a content risk decision result corresponding to the real-time video stream data is determined.

4. The video content risk detection method according to claim 3, wherein determining the content risk decision result corresponding to the real-time video stream data according to each of the preliminary risk decision results and the intermediate risk decision results comprises: Obtaining multimodal risk detection weights corresponding to the first information density data and the second information density data; The preliminary risk decision result and the intermediate risk decision result are integrated according to the multimodal risk detection weight to determine the content risk decision result corresponding to the real-time video stream data.

5. The video content risk detection method according to claim 4, wherein the step of obtaining the multimodal risk detection weights corresponding to the first information density data and the second information density data comprises: Obtaining initial risk detection weights corresponding to the first information density data and the second information density data; According to the content risk decision results of the historical moments associated with the real-time video stream data, the initial risk detection weight is updated to obtain a multimodal risk detection weight, so that the updated multimodal risk detection weight is adapted to the video content of the real-time video stream data.

6. The video content risk detection method according to claim 1, wherein the step of extracting multimodal detection data corresponding to the real-time video stream data comprises: Obtaining a preset detection and extraction time, wherein the detection and extraction time includes a first extraction time length and a second extraction time length, wherein the second extraction time length is greater than the first extraction time length; Extracting the first information density data by combining the first extraction duration with the timestamp information of the real-time video stream data; The second information density data is extracted by combining the second extraction duration with the timestamp information of the real-time video stream data.

7. The video content risk detection method according to claim 6, wherein the first extraction duration includes a video frame capture interval and an audio slice extraction duration; and the first information density data is extracted by combining the first extraction duration with the timestamp information of the real-time video stream data, comprising: Performing frame extraction processing on the real-time video stream data based on the video frame capture interval to obtain a frame capture image, and determining timestamp information corresponding to the frame capture image according to timestamp information of the real-time video stream data; Performing audio extraction on the real-time video stream data based on the audio slice extraction duration to obtain audio slice data, and determining timestamp information corresponding to the audio slice data according to timestamp information of the real-time video stream data; Performing speech recognition on the audio slice data to determine speech recognition text data, and determining timestamp information corresponding to the speech recognition text data according to the timestamp information corresponding to the audio slice data; The frame-cut image, the audio slice data, the speech recognition text data and the corresponding timestamp information are used as the first information density data.

8. The video content risk detection method according to claim 6, wherein the second extraction duration comprises a video slice extraction duration; and the second information density data is extracted by combining the second extraction duration with the timestamp information of the real-time video stream data, comprising: The real-time video stream data is subjected to video extraction based on the video slice extraction duration to obtain video slice data, and an absolute start timestamp and an absolute end timestamp corresponding to the video slice data are determined according to timestamp information of the real-time video stream data.

9. The video content risk detection method according to claim 1, wherein the first information density data at least includes a frame-cut image, an audio slice data and a speech recognition text data; The detecting the first information density data to determine the risk detection result includes: Performing feature analysis on the frame-shot image to obtain frame-shot image features, and determining a frame-shot detection result based on the frame-shot image features; Performing audio analysis on the audio slice data to obtain musical content features, and determining an audio detection result based on the musical content features; Performing content analysis on the speech recognition text data to obtain keyword features, and determining text detection results based on the keyword features; The frame detection result, the audio detection result and the text detection result are stored as risk detection results corresponding to the first information density data.

10. The video content risk detection method according to claim 1, wherein the step of fusing the first information density data and the risk detection result into the second information density data comprises: Obtaining an absolute start timestamp and an absolute end timestamp corresponding to the second information density data; Matching first information density data within a corresponding timestamp range based on the absolute start timestamp and the absolute end timestamp; According to the timestamp information corresponding to the matched first information density data, the matched first information density data and the corresponding risk detection result are integrated into the detection information associated with the second information density data to obtain the fused second information density data.

11. A video content risk detection device, the device comprising: A multimodal detection data extraction module, used to obtain real-time video stream data, and extract multimodal detection data corresponding to the real-time video stream data, wherein the multimodal detection data includes first information density data and second information density data, and the information density of the second information density data is greater than that of the first information density data; A risk detection module, used to detect the first information density data and determine a risk detection result; A content risk decision module is used to fuse the first information density data and the risk detection result into the second information density data, and perform content decision on the fused second information density data to obtain a content risk decision result corresponding to the real-time video stream data.

12. A 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 10 are implemented.

13. An electronic device, characterized in that: include: A processor and a memory; wherein the memory stores a computer program, and the computer program is suitable for being loaded by the processor and executing the steps of the method as claimed in any one of claims 1 to 10.

14. A computer program product having at least one instruction stored thereon, characterized in that: When the at least one instruction is executed by the processor, the steps of the method described in any one of claims 1 to 10 are implemented.

Images