A dual-source approach for 3D human pose estimation from single images

Iqbal, Umar; Doering, Andreas; Yasin, Hashim; Krüger, Björn; Weber, Andreas; Gall, Jüergen

doi:10.1016/j.cviu.2018.03.007

Cited by 42 publications

(29 citation statements)

References 67 publications

(199 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Finally, they balance the potentials' influence by learning the model parameters using a variation of an SVM algorithm. A dual-source approach [29] from 2016. trains a complex machine learning model given two inputs -a generated 3D pose space and an annotated 2D image. The image is used to learn a pictorial structures model for 2D pose estimations.…”

Section: Methodsmentioning

confidence: 99%

“…The pose estimation is generally difficult due to unknown location of a human body in an image, unknown number of people in the scene, (self-) occlusions, a variety of environments [71,43,28] and actions [37,36] and diverse body shapes [7,41] and clothes [58]. Reconstruction of a 3D human pose from a single 2D image is particularly challenging due to 2D-to-3D elevation ambiguities [69,29,42,44,67,66,54,49]. To compensate for the lack of information in a single-view, multi-view [2,30,12,59,47] and multi-frame (video stream) [74,76,33,26,56] approaches to 3D human pose estimation have also been proposed.…”

Section: Introductionmentioning

confidence: 99%

“…Both detection-and regression-based approaches use multi-stage refinements [67,44,54,59] to improve the performance. Body priors (for example, constant body part lengths through the frames or human body symmetries [35]) are also proven to be beneficial for the optimization [66,69,29]. Some deep learning models are able to learn the priors without the explicit clues [30,79].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A Review of 3D Human Pose Estimation from 2D Images

Bartol¹,

Bojanic²,

Petković³

et al. 2020

Proceedings of 3DBODY.TECH 2020 - 11th International Conference and Exhibition on 3D Body Scanning and Processing Technologies,

View full text Add to dashboard Cite

Human pose estimation task takes images as input and extracts a set of locations representing the predefined body joints and the sparse connections between the joints, called the body parts. A pose can be estimated from single or multiple frames, in a single (monocular) or multi-view (stereo) setup and for a single person or multiple people in the scene. In this work, we provide an overview of the classic and deep learning-based 3D pose estimation approaches. We also point out relevant evaluation metrics, pose parametrizations, body models, and 3D human pose datasets. Finally, we review stateof-the-art pose estimation results, briefly discuss open problems, and propose possible future research directions.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A Review of 3D Human Pose Estimation from 2D Images

Bartol¹,

Bojanic²,

Petković³

et al. 2020

Proceedings of 3DBODY.TECH 2020 - 11th International Conference and Exhibition on 3D Body Scanning and Processing Technologies,

View full text Add to dashboard Cite

show abstract

“…DensePose [34] has achieved impressive performance in mapping the pixels of humans in 2D images, containing also background and occlusions, to a 3D representation of the human body. Several other studies [35][36][37] tackle the 3D human pose estimation task by training deep network architectures. Another approach to predict occluded parts of a posture is to rely on the parts of the posture that can still be observed.…”

Section: B Human Posture Trackingmentioning

confidence: 99%

Probabilistic Real-Time User Posture Tracking for Personalized Robot-Assisted Dressing

2019

View full text Add to dashboard Cite

Robotic solutions to dressing assistance have the potential to provide tremendous support for elderly and disabled people. However, unexpected user movements may lead to dressing failures or even pose a risk to the user. Tracking such user movements with vision sensors is challenging due to severe visual occlusions created by the robot and clothes. We propose a probabilistic tracking method using Bayesian networks in latent spaces, which fuses robot end-effector positions and force information to enable camera-less and real-time estimation of the user postures during dressing. The latent spaces are created before dressing by modeling the user movements with a Gaussian Process Latent Variable Model, taking the user's movement limitations into account. We introduce a robot-assisted dressing system that combines our tracking method with hierarchical multi-task control to minimize the force between the user and the robot. The experimental results demonstrate the robustness and accuracy of our tracking method. The proposed method enables the Baxter robot to provide personalized dressing assistance in putting on a sleeveless jacket for users with (simulated) upperbody impairments.

show abstract

“…Another work directly addresses the inference of human poses solely from RGB information (e.g., [26,27]). The additional use of depth information is expected to result in more accurate pose estimates.…”

Section: Related Workmentioning

confidence: 99%

Constraint-Based Optimized Human Skeleton Extraction from Single-Depth Camera

Weinmann

et al. 2019

Sensors

View full text Add to dashboard Cite

As a cutting-edge research topic in computer vision and graphics for decades, human skeleton extraction from single-depth camera remains challenging due to possibly occurring occlusions of different body parts, huge appearance variations, and sensor noise. In this paper, we propose to incorporate human skeleton length conservation and symmetry priors as well as temporal constraints to enhance the consistency and continuity for the estimated skeleton of a moving human body. Given an initial estimation of the skeleton joint positions provided per frame by the Kinect SDK or Nuitrack SDK, which do not follow the aforementioned priors and can prone to errors, our framework improves the accuracy of these pose estimates based on the length and symmetry constraints. In addition, our method is device-independent and can be integrated into skeleton extraction SDKs for refinement, allowing the detection of outliers within the initial joint location estimates and predicting new joint location estimates following the temporal observations. The experimental results demonstrate the effectiveness and robustness of our approach in several cases.

show abstract

A dual-source approach for 3D human pose estimation from single images

Cited by 42 publications

References 67 publications

A Review of 3D Human Pose Estimation from 2D Images

A Review of 3D Human Pose Estimation from 2D Images

Probabilistic Real-Time User Posture Tracking for Personalized Robot-Assisted Dressing

Constraint-Based Optimized Human Skeleton Extraction from Single-Depth Camera

Contact Info

Product

Resources

About