Iterative Greedy Matching for 3D Human Pose Tracking from Multiple Views

Tanke, Julian; Gall, Jüergen

doi:10.1007/978-3-030-33676-9_38

Cited by 23 publications

(28 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…IV-C and IV-D), we extend our method to estimate the poses of multiple persons at a time. Person detections are associated across camera views based on the epipolar distance of their joints using the efficient iterative greedy matching proposed by Tanke et al [28]. The rest of the pipeline is then run for each person observed in at least two views to compute 3D poses and feedback.…”

Section: E Multi-person Pose Estimationmentioning

confidence: 99%

Real-Time Multi-View 3D Human Pose Estimation using Semantic Feedback to Smart Edge Sensors

Bultmann¹,

Behnke²

2021

Robotics: Science and Systems XVII

View full text Add to dashboard Cite

We present a novel method for estimation of 3D human poses from a multi-camera setup, employing distributed smart edge sensors coupled with a backend through a semantic feedback loop. 2D joint detection for each camera view is performed locally on a dedicated embedded inference processor. Only the semantic skeleton representation is transmitted over the network and raw images remain on the sensor board. 3D poses are recovered from 2D joints on a central backend, based on triangulation and a body model which incorporates prior knowledge of the human skeleton. A feedback channel from backend to individual sensors is implemented on a semantic level. The allocentric 3D pose is backprojected into the sensor views where it is fused with 2D joint detections. The local semantic model on each sensor can thus be improved by incorporating global context information. The whole pipeline is capable of realtime operation. We evaluate our method on three public datasets, where we achieve state-of-the-art results and show the benefits of our feedback architecture, as well as in our own setup for multi-person experiments. Using the feedback signal improves the 2D joint detections and in turn the estimated 3D poses.

show abstract

Section: E Multi-person Pose Estimationmentioning

confidence: 99%

Real-Time Multi-View 3D Human Pose Estimation using Semantic Feedback to Smart Edge Sensors

Bultmann¹,

Behnke²

2021

Robotics: Science and Systems XVII

View full text Add to dashboard Cite

show abstract

“…Depending on the number of input cameras, 3D human pose estimation methods are divided into a monocular camera for taking single-view video [2,23,31,14,21,10,22,16,38] and multiple cameras for taking multi-view videos synchronously [3,13,4,32,11,26,7,36,39,35].…”

Section: D Human Pose Estimationmentioning

confidence: 99%

“…While in larger indoor/outdoor environments with more people and cameras, most approaches [4,32,7,39] focus on reducing the computation cost while obtaining better performance. Tanke et al [32] utilize a 2D human pose detector to obtain multiple 2D estimated human poses from multiple views and solve the k-partite matching problem using epipolar geometry to build associations among these multiple 2D estimated human poses across multiple views. They thus construct 3D human pose of each person, followed by a greedy algorithm to match and track iteratively across frames.…”

Section: D Human Pose Trackingmentioning

confidence: 99%

“…In the last, associations between the 2D or 3D skeletons are established with those in the next frame in video streams. Leveraging 2D poses, recent studies [4,32,11,7] follow an initialization-and-tracking framework for 3D pose inference. The framework assumes a streaming mode, i.e., the outputs of multi-view 3D poses are obtained on-line per input frame, and the previously generated outputs cannot be altered.…”

Section: Introductionmentioning

confidence: 99%

“…This is often achieved via epi-polar-line distance and/or person reid matching; then, the initial 3D joints are established with multi-view stereo. As for maintaining the flexibility of use, most works [32,4,7,39] do not need fine-tuning on the data collected in the testing environment; pre-trained person reid models are thus not effective enough due to the testing domain shift in the usual. Hence, epi-polar-line distances are mostly used for initializing the 3D poses.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Part-aware Measurement for Robust Multi-View Multi-Human 3D Pose Estimation and Tracking

Chu

Lee

et al. 2021

2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

View full text Add to dashboard Cite

This paper introduces an approach for multi-human 3D pose estimation and tracking based on calibrated multiview. The main challenge lies in finding the cross-view and temporal correspondences correctly even when several human pose estimations are noisy. Compare to previous solutions that construct 3D poses from multiple views, our approach takes advantage of temporal consistency to match the 2D poses estimated with previously constructed 3D skeletons in every view. Therefore cross-view and temporal associations are accomplished simultaneously. Since the performance suffers from mistaken association and noisy predictions, we design two strategies for aiming better correspondences and 3D reconstruction. Specifically, we propose a part-aware measurement for 2D-3D association and a filter that can cope with 2D outliers during reconstruction. Our approach is efficient and effective comparing to state-of-the-art methods; it achieves competitive results on two benchmarks: 96.8% on Campus and 97.4% on Shelf. Moreover, we extends the length of Campus evaluation frames to be more challenging and our proposal also reach well-performed result. The code will be available at https://git.io/JO4KE.

show abstract

Recursive Bayesian Filtering for Multiple Human Pose Tracking from Multiple Cameras

Kwon

Tanke

Gall

2021

Computer Vision – ACCV 2020

View full text Add to dashboard Cite

Iterative Greedy Matching for 3D Human Pose Tracking from Multiple Views

Cited by 23 publications

References 35 publications

Real-Time Multi-View 3D Human Pose Estimation using Semantic Feedback to Smart Edge Sensors

Real-Time Multi-View 3D Human Pose Estimation using Semantic Feedback to Smart Edge Sensors

Part-aware Measurement for Robust Multi-View Multi-Human 3D Pose Estimation and Tracking

Recursive Bayesian Filtering for Multiple Human Pose Tracking from Multiple Cameras

Contact Info

Product

Resources

About