IntelliVix, a safety AI deep tech company, is establishing a real-time detection and response system for emergencies on highways, including tunnels, areas prone to icing, accident-prone zones, and frequently congested sections.
On July 23, IntelliVix announced that it has secured a contract for the 'Public AI CCTV Conversion Project,' organized by the National Information Society Agency (NIA) with the Korea Expressway Corporation as a participating demand agency.
The project aims to convert existing CCTV systems installed on highways into an intelligent AI CCTV system by utilizing domestic neural processing unit (NPU)-based video analysis servers combined with visual language model (VLM) servers.
The project will cover a total of 4,700 channels, including 3,378 channels in highway tunnels, 422 channels in areas prone to icing, 200 channels in accident-prone zones, and 700 channels in frequently congested areas.
The AI safety monitoring system being developed will analyze CCTV footage in real-time to detect emergencies on the road, such as traffic accidents, stalled vehicles, traffic congestion, wrong-way driving, pedestrian appearances, and falling objects. IntelliVix has applied a dual analysis structure that combines its self-developed video analysis core engine VIX2.0, which excels in rapid detection, with Vision AI and VLM, which understands the context of the footage.
In the first phase, Vision AI will analyze CCTV footage to detect movements and anomalies involving vehicles, pedestrians, and falling objects. In the second phase, VLM will re-verify whether the detected footage represents an actual emergency by analyzing the surrounding road environment and the context of the footage. VLM will automatically generate situational reports that can be immediately utilized by monitoring personnel, interpreting accident types, locations, vehicle and pedestrian movement directions, lane occupancy status, and surrounding traffic flow.
However, existing video analysis systems have faced challenges, such as misidentifying normal situations as accidents due to vehicle shadows, lighting changes, weather conditions, and stalled vehicles, or failing to detect actual accidents. IntelliVix explained that the use of VLM for secondary verification has improved these false positive and false negative issues.
The information detected by AI and verified by VLM will be integrated and managed through IntelliVix's AI safety platform, 'GenAMS.' GenAMS will display the locations of vehicles and pedestrians, accident situation reports, movement directions, and lane occupancy status in real-time on a detailed road map, supporting follow-up responses such as dispatching patrol cars, road control, accident management, and detour guidance.
Previously, IntelliVix established a wildfire trajectory prediction system in Hadong and Sancheong counties using domestic NPUs. This system analyzes wildfire monitoring footage along with weather and terrain information using AI to predict the occurrence and spread of wildfires, aiding disaster response. The highway AI CCTV conversion project represents an expansion of domestically verified NPU-based AI technology from the disaster and safety sector to the transportation sector.
Choi Eun-soo, CEO of IntelliVix, stated, 'This project is significant both technically and industrially as it establishes a large-scale AI safety monitoring system using domestic NPUs without replacing existing CCTV infrastructure. We aim to contribute to creating safe highways where citizens can travel with peace of mind by preventing traffic accidents and secondary incidents through AI technology.'
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.