KOLON BENITECH announced on the 21st that its self-developed safety control system "Artificial Intelligence (AI) Vision Intelligence" has been applied to actual construction sites of KOLON GLOBAL.
The final proof of concept (PoC) work for the advancement of KOLON BENITECH "AI Vision Intelligence" was conducted in a consortium form by KOLON BENITECH, KOLON GLOBAL, Rebellion, and Wish. KOLON BENITECH refined massive construction site data from KOLON GLOBAL to complete a multi-structured AI model optimized for the construction environment.
This work was based on establishing site-specific functional requirements and system checks using CCTV footage and operational data provided by KOLON GLOBAL. Rebellion configured a CCTV monitoring server and technology based on a Neural Processing Unit (NPU) using its proprietary AI semiconductor, ATOM, while Wish provided a monitoring solution that outputs both visual and large language model (LLM) information simultaneously.
The current project "AI Vision Intelligence" is notable for significantly enhancing technical capabilities through collaboration with leading domestic AI corporations. KOLON BENITECH completed a hybrid semiconductor-based AI CCTV safety control system that utilizes both GPU and NPU simultaneously in partnership with the leading domestic AI semiconductor company Rebellion. Following training on real-world data at the KOLON BENITECH AI Solutions Center using GPUs (Graphic Processing Units), the data is then distributed to NPU servers powered by Rebellion's AI semiconductors to further enhance inference capabilities. As a result, descriptions of detected hazardous situations on site have become much more detailed and refined.
"AI Vision Intelligence" is expected to significantly increase the prevention rates of accidents such as improper helmet use, entering hazardous areas, the presence of signal personnel, and heavy machinery approach. This is the effect of a multi-model that transforms visual risk information detected on construction sites into more refined linguistic alert messages developed by KOLON BENITECH. This model maximized LLM inference performance using a high-performance library vLLM (Versatile Large Language Model) that boasts a higher processing speed compared to existing models.
KOLON BENITECH plans to commercialize its independently developed safety control system "AI Vision Intelligence" to create a growth opportunity for its "AI Acceleration Service" to popularize AI utilization. The AI research and development (R&D) institutional sector of KOLON BENITECH is developing AI technologies specialized for various industries with the goal of releasing a "multi-modal AI service" that can process text, images, and audio simultaneously.
Lee Seung-jae, Head of Team at KOLON BENITECH AI R&D, noted, "KOLON BENITECH is pioneering a new business area that maximizes customer business efficiency and convenience through popularizing AI technology while enhancing corporate business innovation," adding, "We will expand the AI business ecosystem by linking our technical capabilities with those of leading domestic corporations that have excellent technological power."