The Electronics and Telecommunications Research Institute (ETRI), in collaboration with the University of Washington, participated in an AI traffic video analysis competition held at a world-renowned computer vision academic conference and secured 1st and 2nd place in two divisions respectively. The integrated traffic video understanding technology 'UniTraffic,' which combines traffic video analysis technology accumulated over 30 years with the latest vision-language model (VLM), was leveraged as the core technology.

 

ETRI announced on the 22nd that the joint research team 'UWIPL_ETRI' recorded 1st place overall in the PSI-VQA division for pedestrian crossing intent analysis and 2nd place overall in the FETV division for intersection traffic violation analysis at the 10th AI City Challenge, a sideline event of 'ECCV 2026,' an international academic conference on computer vision held in Malmö, Sweden.

 

The ECCV 'AI City Challenge' is an AI competition in the intelligent traffic and smart city fields that began in 2017, marking its 10th edition this year.

 

The PSI-VQA division is a task requiring AI to infer pedestrian crossing intent, the timing of risk occurrence, and the reasoning basis through a question-and-answer method from vehicle forward-facing camera footage.

Beyond simple object recognition, AI must comprehensively judge behavioral intent and situational causality.

 

UniTraffic, the core technology of this achievement, is designed to process footage from sources with different forms and perspectives—such as ceiling-mounted CCTV, intersection fisheye lens cameras, and vehicle-mounted cameras—through a single VLM-based system.

 

The research involved researcher Sang-won Kim, principal researcher Byung-geun Kim, and senior researcher Kwang-ju Kim from ETRI's AI Infrastructure Research Division, Daegyeong Research Institute, as well as researcher Jianxu Shangguan and Professor Jenq-Neng Hwang from the University of Washington.

 

Woo-jin Byeon, director of ETRI's Daegyeong Research Institute, stated, "This is the result of inheriting intelligent traffic management and video analysis technology accumulated by predecessor researchers over an extended period and advancing it with cutting-edge AI technology through international collaborative research. We will expand practical demonstration in actual traffic fields and collaboration with domestic companies to develop traffic and safety services that citizens can experience firsthand."

 

ETRI plans to expand the technological application scope to intelligent traffic management, smart cities, and autonomous driving based on this achievement and pursue demonstration and commercialization collaboration with domestic companies.