Abstract: In recent years, with the rapid development of live video broadcasting, a large number of live broadcasting platforms have sprung up, but this also brings huge challenges to the supervision of live broadcasting content. Compared with the review of text content, the review of multimedia content is more complex and more resource-intensive. The technical threshold is relatively high, so this has gradually become a bottleneck restricting the healthy development of live broadcast services, especially for some small and medium-sized live broadcasts that have just started. platform.
background
In recent years, with the rapid development of live video broadcasting, a large number of live broadcasting platforms have sprung up, but this also brings huge challenges to the supervision of live broadcasting content. The review of text content and the review of multimedia content are more complex, and the resource consumption is also higher. The technical threshold is relatively high, so this has gradually become a bottleneck restricting the healthy development of live broadcast services, especially for some small and medium-sized live broadcast platforms that have just started.
For content auditing, some platforms still adopt the method of human flesh auditing, employing a large number of live broadcast auditors to find those non-compliant videos from a large number of live broadcast videos, but limited by human energy, they often cannot achieve good results. , and the shortage salary of auditors has also risen in response.
Fortunately, the rapid development of artificial intelligence has brought a turning point to the solution of this problem, and artificial intelligence review + manual review has become the mainstream.
traditional solution
Companies with technical accumulation and manpower often set up their own AI teams to train their own detection models according to their own business scenarios and massive resources. This has the advantage of being able to better control their own scenarios, but there are also Some disadvantages, such as:
- Users need to complete the frame capture work by themselves, which may consume a lot of computing resources.
- Massive image storage and transmission costs
- Due to technical and human reasons, the accuracy of the auditing algorithm is not satisfactory
Platforms that do not yet have self-built AI teams often use third-party image review APIs, but it is difficult to fine-tune the algorithm, and the review of images does not have the ability to comprehensively judge the context.
Live Audit Solutions
Alibaba Cloud live video auditing can perfectly solve the above problems. There is no high technical threshold. Users can quickly enable live streaming auditing services in the console to solve the pain points of content auditing on small and medium-sized live streaming platforms.
Live Auditing Architecture
- The anchor pushes the stream to the live video center, triggering the live review task
- Framing the live stream at a specified frequency
- Real-time comprehensive content review through the content review engine for screenshots
- The result is handed over to the management and control service for a comprehensive judgment, whether it is necessary to notify the customer and save the evidence frame
Live Auditing Features
low overhead
Transparent to users, users can easily open the video live auditing service on the console without paying separate storage, traffic and computing costs. Users can quickly conduct problems and suspected videos by building their own or using the live auditing platform provided by us. Stream review and ban.
Powerful algorithm support
Relying on Alibaba's years of video and AI technology accumulation, it can quickly and easily access the industry's leading live broadcast review algorithm, and use multi-frame comprehensive judgment to reduce the false detection rate.
Multiple scene support
- Intelligent yellow identification
- Intelligent violence detection
- Ad detection
- Meaningless live broadcast detection (for long-term streaming without meaningful content)
Detection model customization
Due to the diverse needs of users and different scenarios, live auditing provides a more flexible solution. Users can dynamically assign a different audit strategy to each stream, and the algorithm model can also be different for different streams. This is very necessary because different scenarios are very different. In practice, we found that some live streams have a high false detection rate.
Boxing matches, for example, are often misjudged as pornography, because it's very similar to a small movie scene where most two or three people are done.
Another example is the game scene, because the scene of the game itself is rendered, the texture is somewhat different from the normal image collected by the camera, and many games are designed to have a certain violence in it, so the correct difference between the game and the real scene. It is also very necessary. For example, many scenes are easily mistaken for violence and terror.
Therefore, the algorithm must be subdivided for different streams according to the scene, so as to achieve high accuracy and make live auditing more practical. Alibaba Cloud live video auditing provides algorithm control down to the stream level to better cover these scenarios.
Flexible configuration
It can dynamically use different detection scenarios for different streams, different frame cutting frequencies, and the maximum frame cutting frequency can reach one frame per 1s, so that different monitoring intensities can be used for different video streams, focusing on monitoring high-risk streams, which can be further reduced. The cost.
Useful accessibility features
Several independent pictures in practical applications may not be enough to be used as strong evidence to identify violations. Combined with the real-time recording function of the video live broadcast platform, it is possible to start recording in real time for the streams that have been found to have problems, collect evidence, and facilitate manual review.
voice audit
In some complex scenes, such as political affairs, the main risk may be in words, not in vision; some anchors prefer to output some indecent content in words in order to avoid being caught by video pornography. I have used Alibaba Cloud's speech recognition for voice identification and multi-dimensional live broadcast review.
Overview
Alibaba Cloud live video auditing can easily provide users with the ability to audit live videos, greatly reducing the consumption of human resources and improving audit efficiency. Welcome to the official website of Alibaba Cloud to learn more and apply for activation.
To read more good articles, please scan the following QR code: