A South Korean research team has developed a technology that enables drones to accurately figure out objects that are far away at once. The research team already demonstrated the technology at an international competition. It is likely that the technology will be widely used in various fields such as national defense and disaster response.
A research team led by Moon Sung-tae, who is a senior researcher at the Korea Aerospace Research Institute’s Artificial Intelligence Laboratory announced on Monday that it has developed such technology called “DroneEye”.
Object detection is a technology that utilizes artificial intelligence (AI) and distinguishes what the object is from an image. It is already widely used and is normally used on cars or CCTVs that are installed at particular spots. However, the technology has its limits when it is used on cars or CCTVs as their image shooting angle and range become limited.
However, story completely changes once the technology is applied to a drone. A drone based on the technology is able to freely move around a 3D space and observe an object from afar when it is flying at a high altitude. Observing a place from far away means there will be more objects captured in an image. Although the technology can display a much better detection capability, it needs to be more advanced.
The DroneEye technology is able to make the object detection technology more advanced. It allows a drone to distinguish an object that is kilometers away and detect around 500 objects that are captured in an image. A drone is also able to distinguish an object regardless of an angle. For example, it is able to distinguish a person if it looks at a person from a side angle or from above.
The research team already demonstrated the technology at an international competition. At the “VizDrone” competition that was held last year by the ECCV (European Conference on Computer Vision), which is one of the three major AI societies in the world, the research team took away the first place. A drone based on the DroneEye technology showed an accuracy of 34.57%. Although the percentage seems low, it is considerably a high percentage considering the fact that there were many difficult conditions applied to the competition. The research team beat out top teams around the world.
It has been able to develop the technology by having AI learn 10,000 different objects such as human, car, and others.
It plans to continue to improve the DroneEye technology. While object detection currently takes from a server, it plans to implement “edge computing” that enables drones to perform object detection. It also plans to develop the DroneEye technology as a hardware by utilizing AI semiconductors with the Electronics and Telecommunications Research Institute.
Moon said that the DroneEye technology will be applied to formation flight and that the technology will become very useful in various fields including national defense and disaster response.
Staff Reporter Kim, Youngjoon | kyj85@etnews.com