Domestic researchers and the University of Washington (Univ. The object segmentation and tracking technology for autonomous driving jointly developed by a research team from the University of Washington took first place in the world at an international competition for tracking objects in the field of autonomous driving.

The Electronics and Telecommunications Research Institute (ETRI) announced that it took first place in the 'Video Track' of the 'Object Segmentation and Tracking Technology for Autonomous Driving' international competition, sponsored by Google and held at the world's largest computer vision conference (ICCV) for six days starting from the 11th.

A joint research team from ETRI and the University of Washington proposed a deep learning-based object segmentation and tracking framework and won the video track with world-class pixel-level object tracking accuracy.

This competition involves dividing and tracking multiple objects based on road footage filmed from the perspective of an autonomous vehicle.

ETRI Daegyeong Regional Research Center analyzed video provided by the organizers using an algorithm developed through international joint research and tracked about 20 objects, including roads, walls, traffic lights, buildings, and people.

The research team's technology divides objects into pixel units to recognize their shape and color them.

Therefore, detailed identification and sophisticated tracking of objects are possible.

It is a much more advanced technology compared to the existing method of recognizing and tracking objects using a square frame.

This algorithm includes technology that independently determines whether each pixel is an object and tracks changes in the object's position more accurately.

In addition, they stated that they were able to achieve the competition's highest record by utilizing contrastive learning techniques to more accurately recognize the relationships between objects.

ETRI stated that this technology is specialized in the field of object segmentation and tracking for autonomous vehicles.

In addition, weather, lighting changes, object size, occlusion, distance environment, etc. They explained that they confirmed its superior performance compared to other technologies even in various environments.

Object segmentation and tracking technology is a technology capable of accurately and quickly recognizing the locations of vehicles and pedestrians at intersections or on roads.

If applied to traffic control systems for smart cities in the future, it can enhance safety and enable the integration of various services.

ETRI researchers Kwang-Joo Kim and Byung-Geun Kim from the Daegyeong Regional Research Center participated in this video track competition, and Professor Jenq-Neng Hwang's research team from the University of Washington in the United States also participated.

In conjunction with support for the smart city projects being promoted by Daegu Metropolitan City, the research team has been leading the development of visual information-based real-time traffic information recognition technology, which is a core technology in the transportation and crime prevention sectors, as well as multi-object recognition platform technology for safe cities.

Moon Ki-young, Head of the Daegyeong Regional Research Center at ETRI, said, “This technology is a core technology for smart cities, a key field of the Fourth Industrial Revolution, capable of monitoring traffic volume for smart transportation and enhancing vehicle and pedestrian safety. We will continue to conduct related research to apply it to local governments and lead AI-based smart city technology in the future.”