Institutional Repository of Vilnius University
Temporal-aware stereo vision for robotic perception in dynamic environments /
Abstract
dc:descriptionDepth perception is essential for robotic vision, supporting tasks such as navigation and obstacle avoidance. This thesis investigates stereo and monocular depth estimation for robotic perception in dynamic environments and presents a modular framework that supports both approaches. Stereo depth is implemented using classical computer vision methods, including feature matching, epipolar geometry, rectification, and disparity computation (StereoBM) to produce metric depth. Monocular depth estimation is implemented using the MiDaS model to generate relative depth maps from a single RGB image. A Streamlit-based interface enables interactive testing, visualization, and export of results. Qualitative evaluation shows that stereo vision performs well in textured and well-lit scenes but degrades in low-texture regions, while MiDaS remains robust in challenging visual conditions but lacks absolute scale. The findings highlight the complementary nature of stereo and monocular methods for robotic depth perception.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Kumar, Nehal,
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- eng
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD277688945&prefLang=en_US