Abstract: Artificial intelligence is becoming mature in the field of vision. Video content analysis based on artificial intelligence can fundamentally solve the problem of low performance of traditional content analysis methods, and video analysis opens the era of 2.0 intelligence.
As video data volume surges, data processing and content operation costs remain high
Cloud computing, big data, Internet of Things, artificial intelligence and other rapid development of information technology and the transformation of traditional industry digitalization, on the one hand, the amount of media data shows geometric growth, according to IDC forecast, the total global data is expected to reach 44ZB in 2020, China’s data will reach 8060EB, accounting for 18% of the total global data; On the other hand, the production, dissemination and consumption of media have been upgraded. Seventy percent of the vast amount of data will be stored and distributed in the form of pictures and videos. These data from production, dissemination to user consumption, the traditional way of operation cost is high. Artificial intelligence is becoming mature in the field of vision. Video content analysis based on artificial intelligence can fundamentally solve the problem of low performance of traditional content analysis methods. Aided by AI, video analysis will fully enter the era of intelligence.
The whole process of video from production to consumption is intelligent
The whole process of video upgrading from production to consumption is shown in Figure 1.
Figure 1 Video production, editing, distribution, and consumption process upgrades
Huawei cloud EI power video analysis into an intelligent era.
production
- Video collection, from professional equipment to mobile terminals, video content collection is popular. Huawei CLOUD EI can provide special effects processing technologies such as beauty and personalization in the collection stage to facilitate video collection and processing.
- Content review: Before users collect content, upload and store it, Huawei EI cloud provides intelligent review capabilities for video content, including video quality assessment, pornography/violence/terrorism identification, duplicate/similar identification, and copyright fingerprint analysis, so as to prevent duplicate/suspected unsafe content from entering the dissemination process.
The editor
- Structured analysis: Huawei CLOUD EI provides intelligent structured analysis for short and long videos of various types. Videos are described in structured forms such as shots and content fragments, facilitating subsequent content analysis, storage, and distribution.
- Content analysis: Huawei CLOUD EI intelligently analyzes structured video content from scene recognition, subject recognition, behavior/event detection, voice/text analysis, and content description, and supports multi-maintenance content query and retrieval.
- Intelligent editing, based on intelligent content analysis, initially provides the following intelligent video editing technologies:
(1) Video stripping
Compared with long videos, short videos are more conducive to Internet dissemination, and it is indispensable to strip traditional media programs. Huawei Cloud EI can quickly and accurately segment long videos into segments with different themes based on multi-modal information analysis technologies such as characters, scenes, voice and OCR subtitles. Facing mass multimedia videos, it greatly improves the efficiency of video recognition, editing, retrieval and other processing, and reduces the cost of manual operation. The overall efficiency is 10+ times higher than that of manual operation. The accuracy rate of news videos is more than 95%, and the accuracy rate of specific scenes is more than 98%.
(2) Video summary
Static summary: Based on video shot transformation and content understanding technology, Huawei CLOUD EI provides static summary service for different types of long and short videos. It can compress 1 hour video into 5 minutes of summary content, which can be used for quick browsing of content by operation management personnel and quick playing of content by users.
Video concentration: Using background modeling, subject detection, motion track tracking, and front/background fusion technologies, Huawei CLOUD EI provides video concentration services to extract events from extremely long videos and condense them into short video clips for quick viewing of events in video content.
(3) Video cover
Based on video shot transformation, content understanding and analysis technology, it can quickly extract key frames with relevant content and beautiful composition as covers to attract users to watch further. Compared with manual cover selection, the speed of intelligent cover selection is increased by 20+ times and the operating cost is greatly reduced. At the same time, the wonderful cover can attract users’ attention and greatly increase the click rate of videos.
(4) Video content label
Video OCR, video object recognition and video character recognition extract video content from multi-dimensions, transform unstructured video files into structured data, and provide essential information for video search, video recommendation and video operation.
(5) Watermarking
In addition to newly collected videos, a large part of the content uploaded by users is the re-transmission of existing videos. And the repeated transmission of this part of the video is likely to have hit a different watermark. To facilitate enterprise content management, Huawei CLOUD EI provides watermark detection and processing technology, which can intelligently detect watermark locations, reconstruct watermark, and process watermark.
(6) Audio and video fingerprint
Video fingerprint has great application scenarios, including :(1) repeated detection of the content uploaded by users; (2) Copyright tracking can be carried out on the content; (3) Support multi-dimensional and multi-granularity content storage and retrieval;
Huawei cloud provides audio and video fingerprint generation, storage, and retrieval technologies based on audio and video content analysis technologies.
(7) Automatic singing
In the fields of news media and education, there are usually only videos and corresponding audio information. The auteur technology can automatically add subtitles according to videos. Compared with adding subtitles manually, it can greatly improve the timeliness of video release and production efficiency.
- Distribution: Based on the above video structure and intelligent analysis, video content can be stored in a multi-dimensional manner. Combined with user behavior log analysis, it supports fast retrieval and associated recommendation, and pushes the content to the corresponding users at the fastest speed.
- Consumption: In the process of video consumption, Huawei CLOUD EI provides automatic comments and interactive playback technologies to improve end user experience.
Artificial intelligence is entering the industrial production stage characterized by standardization, automation and modularization. AI will bring human society into the age of intelligence. This paper makes a technical analysis on how to realize intelligence in the whole process of video production, editing, distribution and consumption, hoping to provide guidance for friends who are interested in AI and also open up a situation for you who have been suffering from traditional operation mode.
Click to follow, the first time to learn about Huawei cloud fresh technology ~