Single-Stage is Enough: Multi-Person Absolute 3D Pose Estimation

Jin, Lei; Xu, Chenyang; Wang, Xiaojuan; Xiao, Yabo; Guo, Yandong; Nie, Xuecheng; Zhao, Jian

doi:10.1109/cvpr52688.2022.01274

Cited by 23 publications

(9 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The 3D temporal assignment problem was also tackled by the Hungarian matching approach for videobased multi-person 3D HPE, which achieved impressive results in [23], [24]. L Jin et al introduced a single-stage method that integrates human detection and pose estimation, simplifying the process and enhancing efficiency by directly estimating 3D poses from detected individuals in a single network pass, demonstrating significant improvements over traditional multi-stage methods [35].…”

Section: ) Multi-person 3d Hpementioning

confidence: 99%

Evaluating the Accuracy of Cloud-based 3D Human Pose Estimation Tools: A Case Study of MOTiO by RADiCAL

Khalloufi,

Zaifri,

Benlahbib

et al. 2024

IJACSA

View full text Add to dashboard Cite

The use of 3D Human Pose Estimation (HPE) has become increasingly popular in the field of computer vision due to its various applications in human-computer interaction, animation, surveillance, virtual reality, video interpretation, and gesture recognition. However, traditional sensor-based motion capture systems are limited by their high cost and the need for multiple cameras and physical markers. To address these limitations, cloud-based HPE tools, such as DeepMotion and MOTiON by RADiCAL, have been developed. This study presents the first scientific evaluation of MOTiON by RADiCAL, a cloud-based 3D HPE tool based on deep learning and cloud computing. The evaluation was conducted using the CMU dataset, which was filtered and cleaned for this purpose. The results were compared to the ground truth using two metrics, the Mean per Joint Error (MPJPE) and the Percentage of Correct Keypoints (PCK). The results showed an accuracy of 98 mm MPJPE and 96% PCK for most scenarios and genders. This study suggests that cloud-based HPE tools such as MOTiON by RADiCAL can be a suitable alternative to traditional sensorbased motion capture systems for simple scenarios with slow movements and little occlusion.

show abstract

Section: ) Multi-person 3d Hpementioning

confidence: 99%

Evaluating the Accuracy of Cloud-based 3D Human Pose Estimation Tools: A Case Study of MOTiO by RADiCAL

Khalloufi,

Zaifri,

Benlahbib

et al. 2024

IJACSA

View full text Add to dashboard Cite

show abstract

“…More recent studies have further refined such methods. For example, [7] proposed a deep regression network to estimate absolute 3D coordinates directly from images without intermediate 2D pose representations, and [8] improved the accuracy of poste estimations using a keypoint coordinate representation based on camera line-of-sight. Unlike these approaches, which estimate 3D poses from a single image, methods using multiple images of video frames have also been studied; for example, the authors of [9] used temporal convolutional networks (TCNs) to obtain absolute 3D human poses.…”

Section: Related Workmentioning

confidence: 99%

A Simple Method to Calculate Positions in Pose Tracking to Verify Work Procedures

Tanaka

2023

ESL

View full text Add to dashboard Cite

Although manufacturing processes are becoming increasingly automated, many factories still rely on manual operations. In such facilities, there is a strong need to automatically detect human errors by checking whether specified procedures are being followed. For this purpose, studies have been conducted on utilizing deep neural network (DNN) based three-dimensional (3D) human pose-tracking methods to examine work procedures. However, most of these techniques require a high-end computer equipped with a graphics processing unit (GPU). On the other hand, in this study, we adopt MediaPipe Pose, a lightweight pose estimation network provided by Google, to perform pose tracking on a low-end personal computer (PC) to enable such systems to be deployed in small factories. However, MediaPipe Pose cannot track the location of a human body because it estimates poses in a coordinate system with the waist as the origin (that is, in a root‑relative coordinate system). Therefore, in this study, we developed a method to obtain the absolute coordinates of the root with a simple calculation. The results of an experimental evaluation show that the computational load of the proposed approach is negligible, and the repeatability of the estimation sufficed to evaluate a given operator's work on a predetermined working path. Therefore, the proposed methods enables work procedures to be checked using MediaPipe Pose on a low-end PC.

show abstract

“…As 2D pose estimation is maturing, most top-down methods focus on how to regress the depth coordinates of human joints. While bottomup approaches [6,10,40,41] encode the human pose into a representation that does not depend on the number of people. Reference [40] encodes the 3D pose into a 3D volume and refines it using fine-tuned structures.…”

Section: Multi-person 3d Pose Estimationmentioning

confidence: 99%

“…By using depth-aware part association, [41] can reconstruct 3D poses with the constraint of adaptive bone length and ordinal prior. Reference [10] regresses the 2D pose based on the centre point, root-point depth, and root-relative depth. Then it outputs the 3D pose through a single-stage CNN.…”

Section: Multi-person 3d Pose Estimationmentioning

confidence: 99%

MMDA: Multi‐person marginal distribution awareness for monocular 3D pose estimation

Liu

Shuai

et al. 2023

IET Image Processing

View full text Add to dashboard Cite

Most existing 3D pose representations cannot completely decouple the overlapping two or more human joints of the same type. In this paper, the authors propose a novel 2.5 D representation of the human pose by projecting human joints in 3D space onto the three orthogonal planes. The authors apply for the first time the permutation module to a multi‐person 3D human pose estimation task and use Geometric Constraints Loss (GCL) to guide the learning of the model. The authors overcome the negative effects of the inductive bias of convolutional neural networks (CNNs) by aligning the intermediate feature space with the output feature space. The effectiveness of the authors’ approach is validated on the carnegie mellon university (CMU) panoptic dataset and MuPoTS‐3D dataset. The authors’ proposed representations can effectively decouple the human joints in their selected data from overlapping human joints.

show abstract

Single-Stage is Enough: Multi-Person Absolute 3D Pose Estimation

Cited by 23 publications

References 27 publications

Evaluating the Accuracy of Cloud-based 3D Human Pose Estimation Tools: A Case Study of MOTiO by RADiCAL

Evaluating the Accuracy of Cloud-based 3D Human Pose Estimation Tools: A Case Study of MOTiO by RADiCAL

A Simple Method to Calculate Positions in Pose Tracking to Verify Work Procedures

MMDA: Multi‐person marginal distribution awareness for monocular 3D pose estimation

Contact Info

Product

Resources

About