Hanyang University team's AI privacy tech recognized at top global conference

Researchers at Hanyang University have pioneered a groundbreaking technology called "DCL (Dual Convergent Lines)" that fundamentally blocks privacy issues arising from image data in camera-based localization. This innovation was presented at the highly esteemed 'CVPR 2026' (Conference on Computer Vision and Pattern Recognition), a top-tier global academic forum for computer vision and AI.

The DCL technology earned a dual distinction at CVPR 2026, being selected for both an oral presentation and a nomination for the Best Paper Award. Out of 16,092 submissions, only 4,089 were accepted, with just 141 selected for oral presentations and a mere 74 nominated for Best Paper. The Hanyang University team's achievement places their research within the top 1.8% of accepted papers, highlighting its significant impact.

“

This research is a core technology that simultaneously satisfies the accuracy of visual localization and user privacy protection.

Visual localization, a core technology for augmented reality (AR), autonomous driving, and robotics, typically works by extracting feature information from images and comparing it to a 3D map. However, existing systems posed a privacy risk: even if only feature information was transmitted, attackers could potentially reconstruct sensitive original scene data. The DCL technology overcomes this by transforming key image points into a "dual convergent lines" representation based on fixed anchors, rather than transmitting original coordinates. This method neutralizes reconstruction attacks by guiding neighboring lines towards the anchors, preventing attackers from recovering the original key points and thus protecting user privacy.

Extensive testing in various indoor and outdoor environments confirmed DCL's effectiveness. The technology not only completely defends against privacy attacks but also maintains high performance and efficiency for practical visual localization systems. Professor Hong Je-hyung stated, "This research is a core technology that simultaneously satisfies the accuracy of visual localization and user privacy protection." He envisions DCL evolving into "privacy-preserving Spatial AI" applicable to everyday technologies like AR navigation, robot control, and smart cities.

“

We will continue to advance the DCL technology into 'privacy-preserving Spatial AI' applicable to various fields in daily life such as AR navigation, robot control, autonomous driving, digital twins, and smart cities.