Technologies of object segmentation and tracking for autonomous driving jointly developed by the Korean research team and the research team of the University of Washington won the first place prize in an International competition of tracking objects in the field of autonomous driving.
The Electronics and Telecommunications Research Institute (ETRI) announced that it took first place in the international competition hosted by International Conference on Computer Vision (ICCV). It held for 6 days from October 11th. In the competition, a lot of leading institutions in the world, such as China's largest search portal Baidu, Germany's Freiburg University, etc., participated.
The joint research team of the ETRI and the University of Washington in the USA won the video track of 'Multi-Target Tracking' competition with the world-class object tracking accuracy in the unit of the pixel.
ETRI analyzed the video provided by the host with the algorithm developed by the international joint research and tracked more than 20 objects such as roads, walls, traffic lights, buildings, people, etc.
In the research team's technology, objects are divided into pixel units to recognize its shapes and colored. Therefore, detailed identification and sophisticated tracking is possible. By far, it is a more advanced technology than the existing technologies of identifying and tracking an object with a rectangular frame.
This algorithm makes a decision by itself on whether each pixel is an object or not, and includes the technology of tracking the change in location of the object more accurately. Moreover, it announced that it could achieve the best record by utilizing the technique of Contrastive Learning to recognize the correlation between objects more accurately.
ETRI announced that this technology is specialized in the fields of object segmentation and tracking. In addition, it is superior to other technologies in diverse environments, such as weather, change in lighting, object size, occlusion, street environment, etc.
The technology can identify the location of vehicles and pedestrians at intersections or on roads. If this technology is applied to the traffic control system for smart-city, it is possible to improve safety and link various services to it.
The research team presented the applied technology at ICCV Benchmarking Multi-Target Tracking workshop. ETRI also took first and third place in the world in the detection field of the International competition for the state-of-the-art traffic monitoring hosted by Advanced Video and Signal-based Surveillance (AVSS) in 2018.
Ki Young Moon, Vice President of ETRI said, "It is a core technology for smart cities, which is a major field of the 4th industrial revolution. We will continue researches related to it, so that we can lead the AI-based smart city technology by implementing it to local governments."
Source: https://www.nst.re.kr/www/index.do