Jeonbuk National University (JBNU) Master's student Yang Yun-seok (supervisor: Professor Sang‑joon Lee)'s paper has been accepted to the European Conference on Computer Vision (ECCV 2026), the world's leading conference in computer vision. The conference is organized by the European Computer Vision Association and will be held in Malmö, Sweden, from September 8–12, 2026.

The accepted paper, titled 'PDF-Omni: Poincaré Dual Disk Distortion Field-based Recurrent Update for Omnidirectional Stereo Matching,' proposes an omnidirectional stereo matching technique to estimate 360-degree depth from multi-fisheye camera images.

The study proposed a dual Poincaré disk distortion field to model the geometric uncertainty in seam regions caused by fisheye lens radial distortion and viewpoint changes. Based on this, it introduced a distortion-aware recurrent update network that adaptively adjusts region-wise receptive fields.

Through experiments, the method demonstrated significant reductions in depth estimation error, particularly in seam regions, across multiple omnidirectional stereo matching benchmark datasets compared with existing methods.

Yang Yun-seok, the Master's student who led the study (supervisor: Professor Sang‑joon Lee), said, "I am honored to present research conducted during my Master's program at ECCV 2026, the world's leading conference in computer vision."

This paper represents a core study for precisely estimating omnidirectional 360-degree depth information using multi-fisheye camera images. It is expected to make a significant contribution to the advancement of perception technologies for autonomous mobile robots, such as service and delivery robots, which require blind-spot–free spatial awareness.

This research was carried out with support from the Institute for Information & Communications Technology Planning and Evaluation (IITP)'s Regional Intelligence Innovation Talent Training Project (IITP-2026-RS-2024-00439292) and the Ministry of Education's 4th-phase BK21 Project.