mlsanityData to 14 Oct 2026

Systematic Comparison of Projection Methods for Monocular 3D Human Pose Estimation on Fisheye Images

arXiv 2506.19747 ↗First posted 24 Jun 2025cs.CV cs.ROcited by 1

Not on the map. Why?

Abstract (arXiv)

Fisheye cameras offer robots the ability to capture human movements across a wider field of view (FOV) than standard pinhole cameras, making them particularly useful for applications in human-robot interaction and automotive contexts. However, accurately detecting human poses in fisheye images is challenging due to the curved distortions inherent to fisheye optics. While various methods for undistorting fisheye images have been proposed, their effectiveness and limitations for poses that cover a wide FOV has not been systematically evaluated in the context of absolute human pose estimation from monocular fisheye images. To address this gap, we evaluate the impact of pinhole, equidistant and double sphere camera models, as well as cylindrical projection methods, on 3D human pose estimation accuracy. We find that in close-up scenarios, pinhole projection is inadequate, and the optimal projection method varies with the FOV covered by the human pose. The usage of advanced fisheye models like the double sphere model significantly enhances 3D human pose estimation accuracy. We propose a heuristic for selecting the appropriate projection model based on the detection bounding box to enhance prediction quality. Additionally, we introduce and evaluate on our novel dataset FISHnCHIPS, which features 3D human skeleton annotations in fisheye images, including images from unconventional angles, such as extreme close-ups, ground-mounted cameras, and wide-FOV poses, available at: https://www.vision.rwth-aachen.de/fishnchips

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Builds on 39 of its 61 references that are in the map. First posted on arXiv 24 Jun 2025.

Builds on

39 of 61 references.

References · 39

  1. 2015Deep Residual Learning for Image Recognitioncited by 52,096
  2. 2014Microsoft COCO: Common Objects in Contextcited by 17,608
  3. 2021Swin Transformer: Hierarchical Vision Transformer using Shifted Windowscited by 12,139
  4. 2019Deep High-Resolution Representation Learning for Human Pose Estimationcited by 1,339
  5. 2021EfficientNetV2: Smaller Models and Faster Trainingcited by 1,135
  6. 2016Stacked Hourglass Networks for Human Pose Estimationcited by 915
  7. 2023Ego-Exo4D: Understanding Skilled Human Activity from First- and Third-Person Perspectivescited by 490
  8. 2017A simple yet effective baseline for 3d human pose estimationcited by 401
  9. 2022DETRs with Collaborative Hybrid Assignments Trainingcited by 274
  10. 2017Single-Shot Multi-Person 3D Pose Estimation From Monocular RGBcited by 195
  11. 2022RTMDet: An Empirical Study of Designing Real-Time Object Detectorscited by 195
  12. 2023RTMPose: Real-Time Multi-Person Pose Estimation based on MMPosecited by 175
27 more
  1. 2019Camera Distance-aware Top-down Approach for 3D Multi-person Pose Estimation from a Single RGB Imagecited by 132
  2. 2024Nymeria: A Massive Collection of Multimodal Egocentric Daily Motion in the Wildcited by 114
  3. 2022Hierarchically Decomposed Graph Convolutional Networks for Skeleton-Based Action Recognitioncited by 85
  4. 2019xR-EgoPose: Egocentric 3D Human Pose from an HMD Cameracited by 79
  5. 2018Mo2Cap2: Real-time Mobile 3D Motion Capture with a Cap-mounted Fisheye Cameracited by 68
  6. 2023EgoHumans: An Egocentric 3D Multi-Human Benchmarkcited by 60
  7. 2022UnrealEgo: A New Dataset for Robust Egocentric 3D Human Motion Capturecited by 57
  8. 2018The Double Sphere Camera Modelcited by 52
  9. 2020SMAP: Single-Shot Multi-Person Absolute 3D Pose Estimationcited by 52
  10. 2022Learning 3D Human Pose Estimation from Dozens of Datasets using a Geometry-Aware Autoencoder to Bridge Between Skeleton Formatscited by 50
  11. 2022LiDAR-aid Inertial Poser: Large-scale Human Motion Capture by Sparse Inertial and LiDAR Sensorscited by 44
  12. 2022Deep Learning for Omnidirectional Vision: A Survey and New Perspectivescited by 38
  13. 2020MeTRAbs: Metric-Scale Truncation-Robust Heatmaps for Absolute 3D Human Pose Estimationcited by 35
  14. 2022Surround-view Fisheye Camera Perception for Automated Driving: Overview, Survey and Challengescited by 34
  15. 2023SportsPose -- A Dynamic 3D sports pose datasetcited by 26
  16. 2016EgoCap: Egocentric Marker-less Motion Capture with Two Fisheye Cameras (Extended Abstract)cited by 24
  17. 2020RAPiD: Rotation-Aware People Detection in Overhead Fisheye Imagescited by 17
  18. 20203D Object Detection from a Single Fisheye Image Without a Single Fisheye Training Imagecited by 13
  19. 2022Dual networks based 3D Multi-Person Pose Estimation from Monocular Videocited by 13
  20. 2022VirtualPose: Learning Generalizable 3D Human Pose Models from Virtual Datacited by 10
  21. 2022Dynamic Graph Reasoning for Multi-person 3D Pose Estimationcited by 9
  22. 2020Metric-Scale Truncation-Robust Heatmaps for 3D Human Pose Estimationcited by 8
  23. 2022Enhancing Egocentric 3D Pose Estimation with Third Person Viewscited by 5
  24. 2022A survey of top-down approaches for human pose estimationcited by 3
  25. 2022Permutation-Invariant Relational Network for Multi-person 3D Pose Estimationcited by 3
  26. 2024NToP: NeRF-Powered Large-scale Dataset Generation for 2D and 3D Human Pose Estimation in Top-View Fisheye Imagescited by 2
  27. 2025Acquisition of high-quality images for camera calibration in robotics applications via speech promptscited by 1

Built on by

Cited by year

Citing papers per year: 2025: 1.

1 paper.

  1. 2025How do Foundation Models Compare to Skeleton-Based Approaches for Gesture Recognition in Human-Robot Interaction?cited by 1