ETRI Develops Next-Generation Media Technology for AI, Not Humans
TECHWORLD · · 2 views
✦ AI Summary
ETRI will unveil 3 core next-generation AI-based media technologies at KES 2026, including video coding technology for AI.
Unlike video compression for human viewing, this technology efficiently compresses and transmits information needed for AI object detection, classification, and tracking.
ETRI will also unveil 3D human generation and free-viewpoint rendering technology, as well as generative AI-based previsualization video production support technology.
As environments in which AI directly interprets video continue to spread, there is growing demand for technologies different from conventional video compression designed for human viewing. Korean researchers are unveiling a next-generation video compression technology optimized for AI's understanding and analysis of video rather than human viewing. The technology targets autonomous vehicles, robots, drones, and intelligent CCTV, and is designed to efficiently deliver large volumes of video data for direct AI analysis at a low data load.
The Electronics and Telecommunications Research Institute (ETRI) will participate in KES 2026, which runs for 4 days from the 13th. ETRI plans to unveil 3 core next-generation AI-based media technologies, including its "video coding technology for AI."
Among them, the "video coding technology for AI" is distinct from conventional video compression technology for human viewing. The technology is designed to efficiently extract the information needed for AI object detection, classification, and tracking.
It is also a next-generation video coding method that compresses and transmits only the necessary information. ETRI said it plans to unveil these 3 technologies at KES 2026.
Conventional video compression technologies have focused on reducing data volume while preserving perceived image quality for humans as much as possible. By contrast, machine-oriented video coding (VCM) focuses on maintaining AI video analysis performance while efficiently delivering the information AI needs. ETRI has developed related technologies based on the international AI coding standard MPEG VCM.
Rather than transmitting the entire video in the conventional way, this technology efficiently compresses and transmits feature information needed for AI analysis. ETRI explained that this approach can reduce network burden while still securing AI video analysis performance.
ETRI expects the technology to be used in fields such as vehicle-to-vehicle video and information exchange for autonomous driving, intelligent video security, smart cities, drones, and industrial robots. These fields all require real-time processing of large-scale video data.
In line with the spread of physical AI, in which robots and AI increasingly perceive and act in the real world, ETRI expects the technology's potential as a core method for efficiently transmitting video information between machines to grow.
At the KES 2026 venue, ETRI plans to conduct a demonstration in which AI analyzes actual video, extracts feature information, and then efficiently compresses and transmits data based on the extracted information. ETRI said it plans to show the process at KES, where AI analyzes video, pulls out key features, and uses them to compress and transmit data more efficiently.
ETRI will also develop and unveil another technology. Titled "AI-based live-action multi-person 3D human generation and free-viewpoint rendering," it is a technology that uses only footage captured by a small number of multi-view cameras to reconstruct the 3D shapes of multiple people with AI and allow users to watch immersive video from the viewpoint they want.
For this technology, ETRI applied a 3D geometry pretraining encoder and also applied subpixel geometry learning technology. It also enables the rapid generation of high-resolution 3D video with a small number of cameras, implements a forward inference approach that eliminates the need for separate optimization by scene, and delivers high-resolution rendering and fast processing speed.
ETRI said these technologies can rapidly generate high-resolution 3D video even with a small number of cameras. It also explained that the forward inference approach enables both high-resolution rendering and fast speed without separate scene optimization, and that the technology can be used in a wide range of spatial media fields, including sports broadcasting, performances and exhibitions, and extended reality (XR) content.
ETRI has unveiled a "generative AI-based previsualization video production support technology." The technology is designed for the pre-production stage of content such as films, dramas, and commercials, and is intended to help producers realize the scenes they envision in 3D space before actual production begins.
The technology uses sketches, storyboards, and scripts as input analysis targets. AI analyzes multimodal data in various forms, including sketches, storyboards, and scripts, to understand the producer's intent.
AI then automatically converts the producer's intent into an editable, structured 3D scene based on USD (Universal Scene Description). It also uses generative AI to automatically create and place 3D assets such as people, objects, and spaces. Based on this, producers can directly edit and review scene composition, camera positions, and spatial arrangements before actual filming for films, OTT, commercials, and more.
ETRI said that, on the occasion of KES 2026, it plans to promote the intelligence of the media industry across content production, conversion, transmission, and utilization based on the convergence of AI and media technologies.
The direction of the research push includes expanding technological cooperation with domestic companies, increasing commercialization opportunities, and promoting the international standardization of related technologies, with the aim of securing next-generation media technology competitiveness in the AI era.
Lee Tae-jin, head of ETRI's Media Research Division, said that changes in human-centered video compression technology are needed as AI becomes a major consumer of video. He also said ETRI aims to lay the foundation for the efficient transmission and analysis of large-scale video data through video coding technology for AI.
Lee Tae-jin, head of ETRI's Media Research Division, said that in addition to video coding, ETRI will continue AI-media convergence research, including 3D human and generative AI-based content production technologies. He added that through this research, the institute aims to contribute to strengthening competitiveness in fields such as autonomous driving, robotics, smart cities, industrial automation, and next-generation media.
Source: TECHWORLD · Lee Gwang-jae
Original: https://www.epnc.co.kr/news/articleView.html?idxno=407997
References
This article was produced with the help of an automated content generation algorithm.
Source: TECHWORLD
View originalThis article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.