Novel View Synthesis with Neural Radiance Fields for Industrial Robot Applications

Publikation: Beitrag in FachzeitschriftKonferenzaufsatz in FachzeitschriftForschungPeer-Review

Autoren

  • Markus Hillemann
  • Robert Langendörfer
  • Max Heiken
  • Max Mehltretter
  • Andreas Schenk
  • Martin Weinmann
  • Stefan Hinz
  • Christian Heipke
  • Markus Ulrich

Externe Organisationen

  • Karlsruher Institut für Technologie (KIT)
Forschungs-netzwerk anzeigen

Details

OriginalspracheEnglisch
Seiten (von - bis)137-144
Seitenumfang8
FachzeitschriftInternational Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives
Jahrgang48
Ausgabenummer2
PublikationsstatusVeröffentlicht - 11 Juni 2024
VeranstaltungISPRS TC II Mid-term Symposium on the Role of Photogrammetry for a Sustainable World - Las Vegas, USA / Vereinigte Staaten
Dauer: 11 Juni 202414 Juni 2024

Abstract

Neural Radiance Fields (NeRFs) have become a rapidly growing research field with the potential to revolutionize typical photogrammetric workflows, such as those used for 3D scene reconstruction. As input, NeRFs require multi-view images with corresponding camera poses as well as the interior orientation. In the typical NeRF workflow, the camera poses and the interior orientation are estimated in advance with Structure from Motion (SfM). But the quality of the resulting novel views, which depends on different parameters such as the number and distribution of available images, the accuracy of the related camera poses and interior orientation, but also the reflection characteristics of the depicted scene, is difficult to predict. In addition, SfM is a time-consuming pre-processing step, and its robustness and quality strongly depend on the image content. Furthermore, the undefined scaling factor of SfM hinders subsequent steps in which metric information is required. In this paper, we evaluate the potential of NeRFs for industrial robot applications. To start with, we propose an alternative to SfM pre-processing: we capture the input images with a calibrated camera that is attached to the end effector of an industrial robot and determine accurate camera poses with metric scale based on the robot kinematics. We then investigate the quality of the novel views by comparing them to ground truth, and by computing an internal quality measure based on ensemble methods. For evaluation purposes, we acquire multiple datasets that pose challenges for reconstruction typical of industrial applications, like reflective objects, poor texture, and fine structures. We show that the robot-based pose determination reaches similar accuracy as SfM in non-demanding cases, while having clear advantages in more challenging scenarios. We also report results of the novel view quality for different NeRF approaches, showing that an additional online pose refinement may be disadvantageous. Finally, we present first results of applying the ensemble method to estimate the quality of the synthetic novel view in the absence of a ground truth.

ASJC Scopus Sachgebiete

Zitieren

Novel View Synthesis with Neural Radiance Fields for Industrial Robot Applications. / Hillemann, Markus; Langendörfer, Robert; Heiken, Max et al.
in: International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives, Jahrgang 48, Nr. 2, 11.06.2024, S. 137-144.

Publikation: Beitrag in FachzeitschriftKonferenzaufsatz in FachzeitschriftForschungPeer-Review

Hillemann, M, Langendörfer, R, Heiken, M, Mehltretter, M, Schenk, A, Weinmann, M, Hinz, S, Heipke, C & Ulrich, M 2024, 'Novel View Synthesis with Neural Radiance Fields for Industrial Robot Applications', International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives, Jg. 48, Nr. 2, S. 137-144. https://doi.org/10.48550/arXiv.2405.04345, https://doi.org/10.5194/isprs-archives-XLVIII-2-2024-137-2024
Hillemann, M., Langendörfer, R., Heiken, M., Mehltretter, M., Schenk, A., Weinmann, M., Hinz, S., Heipke, C., & Ulrich, M. (2024). Novel View Synthesis with Neural Radiance Fields for Industrial Robot Applications. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives, 48(2), 137-144. https://doi.org/10.48550/arXiv.2405.04345, https://doi.org/10.5194/isprs-archives-XLVIII-2-2024-137-2024
Hillemann M, Langendörfer R, Heiken M, Mehltretter M, Schenk A, Weinmann M et al. Novel View Synthesis with Neural Radiance Fields for Industrial Robot Applications. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives. 2024 Jun 11;48(2):137-144. doi: 10.48550/arXiv.2405.04345, 10.5194/isprs-archives-XLVIII-2-2024-137-2024
Hillemann, Markus ; Langendörfer, Robert ; Heiken, Max et al. / Novel View Synthesis with Neural Radiance Fields for Industrial Robot Applications. in: International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives. 2024 ; Jahrgang 48, Nr. 2. S. 137-144.
Download
@article{51804e1a5b3e49b2a19cf86e9cef0900,
title = "Novel View Synthesis with Neural Radiance Fields for Industrial Robot Applications",
abstract = "Neural Radiance Fields (NeRFs) have become a rapidly growing research field with the potential to revolutionize typical photogrammetric workflows, such as those used for 3D scene reconstruction. As input, NeRFs require multi-view images with corresponding camera poses as well as the interior orientation. In the typical NeRF workflow, the camera poses and the interior orientation are estimated in advance with Structure from Motion (SfM). But the quality of the resulting novel views, which depends on different parameters such as the number and distribution of available images, the accuracy of the related camera poses and interior orientation, but also the reflection characteristics of the depicted scene, is difficult to predict. In addition, SfM is a time-consuming pre-processing step, and its robustness and quality strongly depend on the image content. Furthermore, the undefined scaling factor of SfM hinders subsequent steps in which metric information is required. In this paper, we evaluate the potential of NeRFs for industrial robot applications. To start with, we propose an alternative to SfM pre-processing: we capture the input images with a calibrated camera that is attached to the end effector of an industrial robot and determine accurate camera poses with metric scale based on the robot kinematics. We then investigate the quality of the novel views by comparing them to ground truth, and by computing an internal quality measure based on ensemble methods. For evaluation purposes, we acquire multiple datasets that pose challenges for reconstruction typical of industrial applications, like reflective objects, poor texture, and fine structures. We show that the robot-based pose determination reaches similar accuracy as SfM in non-demanding cases, while having clear advantages in more challenging scenarios. We also report results of the novel view quality for different NeRF approaches, showing that an additional online pose refinement may be disadvantageous. Finally, we present first results of applying the ensemble method to estimate the quality of the synthetic novel view in the absence of a ground truth.",
keywords = "industrial applications, neural radiance fields, novel view synthesis, robotics, uncertainty estimation",
author = "Markus Hillemann and Robert Langend{\"o}rfer and Max Heiken and Max Mehltretter and Andreas Schenk and Martin Weinmann and Stefan Hinz and Christian Heipke and Markus Ulrich",
note = "Publisher Copyright: {\textcopyright} Author(s) 2024.; ISPRS TC II Mid-term Symposium on the Role of Photogrammetry for a Sustainable World ; Conference date: 11-06-2024 Through 14-06-2024",
year = "2024",
month = jun,
day = "11",
doi = "10.48550/arXiv.2405.04345",
language = "English",
volume = "48",
pages = "137--144",
number = "2",

}

Download

TY - JOUR

T1 - Novel View Synthesis with Neural Radiance Fields for Industrial Robot Applications

AU - Hillemann, Markus

AU - Langendörfer, Robert

AU - Heiken, Max

AU - Mehltretter, Max

AU - Schenk, Andreas

AU - Weinmann, Martin

AU - Hinz, Stefan

AU - Heipke, Christian

AU - Ulrich, Markus

N1 - Publisher Copyright: © Author(s) 2024.

PY - 2024/6/11

Y1 - 2024/6/11

N2 - Neural Radiance Fields (NeRFs) have become a rapidly growing research field with the potential to revolutionize typical photogrammetric workflows, such as those used for 3D scene reconstruction. As input, NeRFs require multi-view images with corresponding camera poses as well as the interior orientation. In the typical NeRF workflow, the camera poses and the interior orientation are estimated in advance with Structure from Motion (SfM). But the quality of the resulting novel views, which depends on different parameters such as the number and distribution of available images, the accuracy of the related camera poses and interior orientation, but also the reflection characteristics of the depicted scene, is difficult to predict. In addition, SfM is a time-consuming pre-processing step, and its robustness and quality strongly depend on the image content. Furthermore, the undefined scaling factor of SfM hinders subsequent steps in which metric information is required. In this paper, we evaluate the potential of NeRFs for industrial robot applications. To start with, we propose an alternative to SfM pre-processing: we capture the input images with a calibrated camera that is attached to the end effector of an industrial robot and determine accurate camera poses with metric scale based on the robot kinematics. We then investigate the quality of the novel views by comparing them to ground truth, and by computing an internal quality measure based on ensemble methods. For evaluation purposes, we acquire multiple datasets that pose challenges for reconstruction typical of industrial applications, like reflective objects, poor texture, and fine structures. We show that the robot-based pose determination reaches similar accuracy as SfM in non-demanding cases, while having clear advantages in more challenging scenarios. We also report results of the novel view quality for different NeRF approaches, showing that an additional online pose refinement may be disadvantageous. Finally, we present first results of applying the ensemble method to estimate the quality of the synthetic novel view in the absence of a ground truth.

AB - Neural Radiance Fields (NeRFs) have become a rapidly growing research field with the potential to revolutionize typical photogrammetric workflows, such as those used for 3D scene reconstruction. As input, NeRFs require multi-view images with corresponding camera poses as well as the interior orientation. In the typical NeRF workflow, the camera poses and the interior orientation are estimated in advance with Structure from Motion (SfM). But the quality of the resulting novel views, which depends on different parameters such as the number and distribution of available images, the accuracy of the related camera poses and interior orientation, but also the reflection characteristics of the depicted scene, is difficult to predict. In addition, SfM is a time-consuming pre-processing step, and its robustness and quality strongly depend on the image content. Furthermore, the undefined scaling factor of SfM hinders subsequent steps in which metric information is required. In this paper, we evaluate the potential of NeRFs for industrial robot applications. To start with, we propose an alternative to SfM pre-processing: we capture the input images with a calibrated camera that is attached to the end effector of an industrial robot and determine accurate camera poses with metric scale based on the robot kinematics. We then investigate the quality of the novel views by comparing them to ground truth, and by computing an internal quality measure based on ensemble methods. For evaluation purposes, we acquire multiple datasets that pose challenges for reconstruction typical of industrial applications, like reflective objects, poor texture, and fine structures. We show that the robot-based pose determination reaches similar accuracy as SfM in non-demanding cases, while having clear advantages in more challenging scenarios. We also report results of the novel view quality for different NeRF approaches, showing that an additional online pose refinement may be disadvantageous. Finally, we present first results of applying the ensemble method to estimate the quality of the synthetic novel view in the absence of a ground truth.

KW - industrial applications

KW - neural radiance fields

KW - novel view synthesis

KW - robotics

KW - uncertainty estimation

UR - http://www.scopus.com/inward/record.url?scp=85197339144&partnerID=8YFLogxK

U2 - 10.48550/arXiv.2405.04345

DO - 10.48550/arXiv.2405.04345

M3 - Conference article

AN - SCOPUS:85197339144

VL - 48

SP - 137

EP - 144

JO - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives

JF - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences - ISPRS Archives

SN - 1682-1750

IS - 2

T2 - ISPRS TC II Mid-term Symposium on the Role of Photogrammetry for a Sustainable World

Y2 - 11 June 2024 through 14 June 2024

ER -

Von denselben Autoren