cs.CVOct 26, 2026

Zero-Shot Monocular Depth Estimation for Visual Perception in Lower-Limb Prostheses

Authors: Ishi Vijayvergiya, Marc Hölle, Youssef Dawoud, Amir El-Ghoussani, Michele De Vita, Rohan Asthana, Philipp Beckerle, Vasileios Belagiannis

Organizations: Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Monocular depth estimation has advanced rapidly with the development of foundation models that demonstrate strong zero-shot generalisation across diverse visual domains. However, the applicability of these models to visual perception for lower-limb prosthetics remains largely unexplored. Reliable depth perception is particularly important for prosthesis users, who often experience significant challenges when navigating stairs, uneven terrain, and other complex environments. To address this gap, we introduce the Lower-Limb Prosthesis (LP) dataset, a benchmark designed to evaluate depth perception from an emulated knee-level prosthetic viewpoint. Using this dataset, we evaluate four state-of-the-art monocular depth estimation foundation models. Our results provide the first systematic benchmark of monocular depth estimation performance from a prosthetic viewpoint, while highlighting both the challenges and the potential of deploying foundation models in prosthetic perception systems. The code and dataset are available at https://github.com/Ishivijay/lp_dataset.

Figures & tables

Fig. 1: LP dataset acquisition setup. The dataset was collected with a wearable camera mounted on the lower limb in four configurations: above or below the knee, on each leg. These configurations yield egocentric recordings from an emulated knee-level prosthetic viewpoint during locomotion across a diverse set of mounting positions.
SequenceDirectionImages
LA downstairsDown428
LA upstairsUp629
LB upstairsUp591
RA downstairsDown375
RA upstairsUp622
RB downstairsDown499
TABLE I: Overview of the six recording sequences in the LP dataset.
Fig. 2: Qualitative comparison of zero-shot monocular depth estimation on the LP dataset. Nine representative scenes (rows), spanning corridors, staircases, landings, and doorways, are shown with, from left to right, the RGB input, the Ground truth and the depth predicted by Marigold, Depth Anything V2, Depth Pro, and VGGT. Depth is rendered with a colormap, where warm colors denote near and cool colors denote far surfaces. Black regions in the ground truth denote invalid sensor returns that are caused by surfaces beyond the operating range, grazing-angle specular reflections on floor tiles, or areas outside the depth field of view.
MetricMarigoldDA-V2Depth ProVGGT
Abs Rel ↓\downarrow0.15680.20660.15990.1532
Sq Rel ↓\downarrow0.08580.11580.08290.0818
RMSE ↓\downarrow0.35490.39470.34890.3449
RMSE log ↓\downarrow0.19720.28060.20060.2011
δ1\delta_{1} ↑\uparrow0.80750.71790.80080.8099
δ2\delta_{2} ↑\uparrow0.94340.89550.94080.9434
TABLE II: Zero-shot depth estimation results on the LP dataset (Staircase scene) with least-squares alignment. Lower is better for ↓\downarrow metrics; higher is better for ↑\uparrow metrics. Best in bold .
MetricMarigoldDA-V2Depth ProVGGT
Abs Rel ↓\downarrow0.19630.27390.22190.2477
Sq Rel ↓\downarrow0.08130.15130.09360.1241
RMSE ↓\downarrow0.33570.50670.35950.4084
RMSE log ↓\downarrow0.22050.39800.24830.2898
δ1\delta_{1} ↑\uparrow0.75920.54420.66320.6100
δ2\delta_{2} ↑\uparrow0.93710.81190.91840.8824
TABLE III: Zero-shot depth estimation results on the LP dataset (Corridor scene) with least-squares alignment. Lower is better for ↓\downarrow metrics; higher is better for ↑\uparrow metrics. Best in bold .
More in this topic

Monocular Depth Estimation

Monocular depth estimation infers scene geometry and distance from a single camera image, enabling 3D perception without stereo or depth sensors.

See all 135 papers

Explore similar work

CardsList

  1. ZipDepth: Bringing Lightweight Zero-Shot Monocular Depth Anywhere, on Any Device

    Jul 9, 2026Fabio Tosi, Luca Bartolomei, Matteo Poggi +1Efficient Neural Network InferenceMonocular Depth Estimation

    University of Bologna, Bologna, Italy

  2. Depth2Pose: A Pose-Based Benchmark for Monocular Depth Estimation without Ground-Truth Depth

    May 19, 2026Viktor Kocur, Sithu Aung, Gabrielle Flood +4Depth EstimationMonocular Depth Estimation

    Faculty of Mathematics, Physics and Informatics, Comenius University in Bratislava · Visual Recognition Group, Faculty of Electrical Engineering, Czech Technical University in Prague · Czech Institute of Informatics, Robotics and Cybernetics, Czech Technical University in Prague

  3. EndoUFM: Utilizing Foundation Models for Monocular depth estimation of endoscopic images

    Aug 25, 2025Xinning Yao, Bo Liu, Bojian Li +3Medical ImagingMonocular Depth Estimation

    Image Processing Center, Beihang University, Beijing, 100191, China. · State Key Laboratory of High-Efficiency Reusable Aerospace Transportation Technology, Beijing 102206, China.