본문 바로가기
  • Home

A Zero-Shot Vision Algorithm for Safe Path Planning on Irregular Terrain in Unstructured Outdoor Environments

  • Journal of Internet of Things and Convergence
  • Abbr : JKIOTS
  • 2026, 12(4), 25
  • Publisher : The Korea Internet of Things Society
  • Research Area : Engineering > Computer Science > Internet Information Processing
  • Received : June 29, 2026
  • Accepted : August 18, 2026
  • Published : August 31, 2026

Sowon Lee 1 Taehoon Kim 2 AHN JUN HO ORD ID 3

1경기대학교 인공지능전공
2경기대학교 컴퓨터과학전공
3경기대학교

Accredited

ABSTRACT

As the demand for outdoor autonomous mobile robots increases, zero-shot methods that identify safe traversable areas while minimizing dataset construction and additional training are gaining attention. However, existing monocular vision-based approaches often show limited generalization in unstructured outdoor environments with ambiguous boundaries and diverse surface conditions. This study proposes a zero-shot fusion algorithm integrating image segmentation, depth estimation, and feature extraction to distinguish obstacles and hazardous regions without additional training. The segmentation parameters are adjusted to generate dense masks that include small obstacles, while a depth estimation model calculates pixel-wise relative distances. Feature extraction is then used to identify highly activated masks corresponding to traversable paths, and distribution-based cosine similarity and hierarchical clustering merge over-detected regions to improve detection stability. Finally, average depth values are assigned to each mask to incorporate spatial information. The proposed method achieved accuracies of 92.6% on a self-constructed dataset and 95.5% on the ORFD benchmark.

Journal Copyright Policy

No CCL information provided

Citation status

* References for papers published after 2025 are currently being built.