Detecting Insects, Teaching Robots, Analyzing Gait... UNIST Computer Vision Scores a String of Wins

Won first, second, and third place across three categories at international competitions

From detecting insects in farmland to robotic grasping motions

AI technologies that detect small insects in farmland and teach robotic hands how to grasp objects have won awards at international competitions in succession. A technology that analyzes patients' gait to predict the severity of walking impairments also achieved success.

A team led by Professor Baek Seungnyeol of the UNIST Graduate School of Artificial Intelligence, headed by President Chong Rae Park, placed first, second, and third in three categories of international challenges held alongside the European Conference on Computer Vision (ECCV 2026) in Malmö, Sweden, the university announced on October 8.

Computer vision is a technology that recognizes and analyzes objects and people's movements in video footage.

Kim Junsu, an integrated master's and doctoral student, placed first among 37 teams in the "BuzzSpot" challenge, which involved detecting pollinating insects, including bees and hoverflies, in farmland footage.

Kim Junsu, an integrated master's and doctoral student who placed first in the ECCV 2026 "BuzzSpot" Challenge.

Insects in farmland footage are difficult to detect because they are small and often obscured by plants. The research team supplemented training data using image crops of insects for which data was scarce, and trained the AI to distinguish the characteristics of different insect species.

The team also improved performance while reducing computational demands by analyzing each video in a single pass. The technology could also be used to monitor farmland insects and survey biodiversity.

Integrated master's and doctoral students Kim Woojin and Kim Sinu, along with postdoctoral researcher Muneeb Ahmed Khan, placed second among eight teams in the "DexGraspMotion" challenge, which involved generating object-grasping movements for robotic hands.

Kim Woojin (from left), Kim Shinwoo, an integrated master's and doctoral program student, and Munib, a postdoctoral researcher, who placed second in the DexGraspMotion Challenge at ECCV 2026.

Instead of repeatedly teaching the robot a large number of movements, the team applied a method that adapts a single successful demonstration of grasping an object to the object's position and shape. In simulations, the method generated stable grasping movements even for objects the robot had not encountered before.

Ph.D. students Kim Suji and Kim Minkyung, along with postdoctoral researcher Muhammad Munsif, placed third among 58 teams in the "MoCha" challenge, which involved analyzing human movement.

The task was to predict the severity of walking impairments using 3D gait data from patients with Parkinson's disease. The team trained the model on gait movements together with sentences describing them, and combined AI models developed by multiple medical institutions into a single model.

The UNIST research team placed third in the ECCV 2026 “Moka” Challenge. From left: doctoral student Kim Suji, postdoctoral researcher Muhammad Munsif, doctoral student Kim Minkyeong, and Professor Baek Seung-ryeol.

A key feature is that the approach reduces the problem of fitting the data of a particular medical institution alone, enabling the model to maintain its performance on data from other medical institutions that was not used for training.

"This shows that computer vision technology can be used to solve problems in a variety of fields, including agriculture, robotics, and clinical care," Professor Baek Seungnyeol said. "We will continue researching vision AI technologies that can be applied in real-world settings."

This content was produced with the assistance of AI translation services.