CoChat

The DOI matches its Crossref record; 47 of the 50 references it cites were found in a registry or an index; 3 of the 15 sources its claims cite were checked for support.

Keypoint Detection and Description through Deep Learning in Unstructured Environments

Open access

Abstract

Feature extraction plays a crucial role in computer vision and autonomous navigation, offering valuable information for real-time localization and scene understanding. However, although multiple studies investigate keypoint detection and description algorithms in urban and indoor environments, far fewer studies concentrate in unstructured environments. In this study, a multi-task deep learning architecture is developed for keypoint detection and description, focused on poor-featured unstructured and planetary scenes with low or changing illumination. The proposed architecture was trained and evaluated using a training and benchmark dataset with earthy and planetary scenes. Moreover, the trained model was integrated in a visual SLAM (Simultaneous Localization and Maping) system as a feature extraction module, and tested in two feature-poor unstructured areas. Regarding the results, the proposed architecture provides a mAP (mean Average Precision) in a level of 0.95 in terms of keypoint description, outperforming well-known handcrafted algorithms while the proposed SLAM achieved two times lower RMSE error in a poor-featured area with low illumination, compared with ORB-SLAM2. To the best of the authors’ knowledge, this is the first study that investigates the potential of keypoint detection and description through deep learning in unstructured and planetary environments.

Overview

CoChat Summary

TL;DR

The paper develops HF-net2, a multi-task teacher–student deep learning model for keypoint detection and description in unstructured and planetary scenes, and integrates it into an ORB-SLAM2-based visual SLAM system. It reports that HF-net2 outperforms several handcrafted baselines and that the proposed SLAM achieves lower trajectory error than ORB-SLAM2, especially under low illumination.

Key findings

  • For keypoint description, HF-net2 achieved mAP 0.98 under illumination changes and 0.95 under viewpoint changes, with matching scores of 0.74 and 0.63, respectively; this outperformed original HF-net, SIFT, and ORB but remained below teacher SuperPoint (mAP 0.99/0.99).
  • For keypoint detection, HF-net2 achieved repeatability 0.74 and mAP 0.71 under illumination changes, and repeatability 0.69 and mAP 0.49 under viewpoint changes; it followed SuperPoint and outperformed SIFT and FAST in the reported comparisons, though Harris had higher viewpoint repeatability (0.77).
  • In SLAM experiments, the HF-net2-based SLAM had RMSE 0.12 m vs. ORB-SLAM2’s 0.15 m in a rocky low-illumination square path; RMSE 0.21 m vs. 0.34 m in sandy high illumination; and RMSE 0.24 m vs. 0.50 m in sandy artificial low illumination, with standard deviation 0.06 m vs. 0.12 m.
  • The model achieved about 16 ms per-frame inference time and 62.5 FPS on a GPU-enabled machine.

Methods

  • HF-net2 is based on HF-net and uses a multi-teacher–student architecture: SuperPoint is the teacher for keypoint detection and local description, NetVLAD is the teacher for global description, and MobileNetV3-large is the shared encoder with separate keypoint, local descriptor, and global descriptor decoders.
  • Training used 48,000 FPV/rover-based images from Earth, the Moon, and Mars, including 26,000 Earth RGB images, about 13,000 NASA Curiosity Mars images, and about 9,000 artificial lunar rover images; models were trained for 30,000 iterations with RMSProp at learning rates 0.001–0.00001 using TensorFlow on an NVIDIA GeForce GTX 1080 Ti.
  • Evaluation used a benchmark of 120 sequences from Earth, Mars, and the Moon, with 60 illumination-change sequences and 60 viewpoint-change sequences, each containing an original image plus five transformed representations; the SLAM system was based on ORB-SLAM2 with HF-net2 as the feature extractor, tested in rocky and sandy scenes with an Intel RealSense D435 RGB-D camera and RTK GNSS ground truth.

Limitations

  • The paper states future work will further improve and fine-tune the architecture by enriching the training and evaluation dataset with more rover-based data, and will optimize the SLAM loop-closing module using only the global descriptor instead of a BoW algorithm, which can fail in completely unknown environments.
  • Evident from the text: the SLAM validation is limited to two terrestrial rocky and sandy areas with a small set of paths and illumination conditions, not actual planetary deployment.
  • Evident from the text: the benchmark includes artificial lunar images and generated illumination/viewpoint transformations, and HF-net2 generally remains below its teacher SuperPoint in the reported detector and descriptor metrics, except in one qualitative artificial lunar scene.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/robotics12050137

Cites

47 of 47 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

3 Valid · 2 Not valid · 7 Not checked · 62 Own results and interpretations

15 more claims

Create a free account to see all 74 claims and every source verdict.

See all claims — free account
  • Probably not a substantial claim. 28
  • It may state two separate claims as one. 32
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

3 of 3

Citations

Cites

Show all 47 references

Research works printed but not linked: 1

  1. Howard, A., Sandler, M., Chu, G., Chen, L., Chen, B., Tan, M., Wang, W., Zhu, Y., Pang, R., and Vasudevan, V. (November, January 27). Searching for MobileNetV3. Proceedings of the ICCV, Seoul, Republic of Korea.

Cited by

See all 10

Metadata

Authors
Petrakis, G., & Partsinevelos, P.
Affiliations
Technical University of Crete (GR)
Corresponding author
Partsinevelos, Panagiotis
Venue
Robotics
Publisher
MDPI AG
Published
2023-09-30
Volume
12
Issue
5
Pages
137
ISSN
2218-6581 (ISSN-L)
OpenAlex
W4387234764
Cited by
10
References
50
Citations by year
2026: 3 · 2025: 3 · 2024: 4
FWCI
1.21
Type
journal article
Language
en
Open access
Locations
  • Robotics (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • European Commission
  • European Social Fund
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Robotics and Sensor-Based Localization
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Image and Video Retrieval Techniques
  • Physical Sciences › Earth and Planetary Sciences › Geology › 3D Surveying and Cultural Heritage
Keywords
keypoint detection, keypoint description, feature extraction, unstructured environments, low-light conditions, visual SLAM, multi-task deep learning, autonomous navigation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-07, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-07 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Keypoint Detection and Description through Deep Learning in Unstructured Environments. https://app.cochat.ai/papers/10.3390/robotics12050137
The paper
Petrakis, G., & Partsinevelos, P. (2023). Keypoint Detection and Description through Deep Learning in Unstructured Environments. Robotics. https://doi.org/10.3390/robotics12050137

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.