Nota, a company specializing in AI lightweighting and optimization technology, has partnered with KBS to build an AI-based disaster special report video analysis system that automatically analyzes and selects CCTV footage in the event of a disaster. The company explained that it has established a system capable of identifying scenes suitable for broadcast transmission in real time.

Nota announced on the 19th that it has completed KBS's 'Disaster CCTV AI Dataset Construction and Video Analysis Advancement' project and implemented an on-device AI-based disaster news special report workflow utilizing a Vision Language Model (VLM).

Previously, during disaster situations, personnel had to manually review a large volume of CCTV footage, which limited the ability to make quick decisions. In particular, during major disasters such as wildfires and torrential rains, the system was designed so that screening time was required as video footage was received all at once. This project focuses on improving these issues by automatically selecting footage suitable for broadcast.

The core technology is Nota's 'Nota Vision Agent (NVA).' NVA is a solution designed to run inside the device by lightweighting high-performance VLM, recognizing the context within the video and analyzing its meaning. It is a structure that automatically extracts scenes with high news value from footage collected by CCTVs near disaster areas and presents them with priority. The on-device approach minimizes network latency to enable a response to emergency situations.

According to Nota, internal tests using a wildfire dataset showed a high concordance rate between images deemed suitable for breaking news by reporters and those selected by the NVA. They also explained that the system is designed to complete analysis within tens of seconds, even in environments where a large volume of images is received simultaneously, and to present the rationale for each scene.

KBS has introduced a feedback system in which reporters evaluate the suitability of video footage for news reporting, and the results are then reflected back into AI training. This structure accumulates on-site opinions as data to gradually improve the model. It is considered significant for establishing an operational system that takes into account not only technical performance but also the actual broadcast production environment.

The demand for AI-based automation in disaster response is expanding within the broadcasting and media industry. As large-scale video data analysis and rapid decision-making are required, the potential for utilizing on-device AI and VLM-based video understanding technologies is garnering attention.

Nota announced that it plans to expand collaboration in the public and media sectors based on this case. KBS also plans to gradually advance its AI infrastructure applicable to the production environment of special disaster reports.