'Not humans, but AI watches the footage': ETRI develops next-generation video compression technology

South Korean researchers have developed a next-generation video compression technology capable of efficiently processing the vast amounts of visual data generated by self-driving cars, robots and drones.

The Electronics and Telecommunications Research Institute (ETRI) said it will unveil three core next-generation media technologies — including a video coding system designed for AI — at KES 2026, the Korea Electronics Show, which runs for four days starting Tuesday at COEX in Seoul.

The centerpiece of the showcase is what ETRI calls "video coding for machines." Unlike conventional compression methods developed for human viewers, the technology selectively compresses and transmits only the information AI needs to detect, classify and track objects within video footage.

While traditional video compression focuses on reducing file size while preserving picture quality as perceived by humans, video coding for machines (VCM) prioritizes maintaining AI analysis performance while improving data transmission efficiency.

Instead of sending entire video streams through conventional methods, the technology efficiently compresses feature data required for AI analysis, easing the burden on networks. ETRI said it expects the technology to find applications particularly in fields that require real-time processing of large-scale video data — including the exchange of visual information between self-driving cars, intelligent CCTV systems, smart cities, drones and industrial robots.

ETRI will also present a technology that renders multiple people as realistic three-dimensional images using only a small number of cameras.

The AI-based realistic multi-person 3D human generation and free-viewpoint rendering technology uses AI to analyze footage captured from multiple angles, reconstruct a person's three-dimensional form, and allow viewers to watch the resulting video from any desired angle.

The system achieves high-definition rendering and fast processing speeds simultaneously by employing a feed-forward inference method. ETRI said it expects the technology to be applicable across a range of fields, including sports broadcasting, performances, exhibitions and extended reality (XR) content.

ETRI will also demonstrate a technology that integrates generative AI into the production of films, drama series and advertisements.

The generative AI-based pre-visualization production support technology uses AI to analyze sketches, storyboards and scripts, then renders scenes envisioned by creators in three-dimensional space before production begins.

ETRI said it plans to expand technology cooperation and commercialization with domestic companies based on the technologies unveiled at the show, and to take an active role in related international standardization efforts.

"As AI emerges as a major consumer of video content, new changes are being demanded of existing human-centered video compression technology," said Lee Tae-jin, head of ETRI's Media Research Division. "Through video coding for machines, we will lay the groundwork for delivering and analyzing large-scale video data more efficiently."