Article
Computer Science, Interdisciplinary Applications
Daewoon Kim, Kwanghee Ko
Summary: This paper presents a novel deep learning-based camera localization method that improves the accuracy of pose estimation from a single RGB image by using iterative relative pose estimation. The method trains the network for absolute poses and relative poses simultaneously using Siamese networks. Experimental results demonstrate that the proposed method achieves higher localization accuracy than state-of-the-art deep learning-based camera localization methods.
JOURNAL OF COMPUTATIONAL DESIGN AND ENGINEERING
(2022)
Article
Automation & Control Systems
Xin Wu, Yonghui Wang, Lei Chen, Lin Zhang, Lianming Wang
Summary: Motion parameters measurement is important for understanding animal behavior and object motion laws. This paper introduces a method using 3D pose estimation for better accuracy, which was successfully evaluated on various scenarios. The enhanced bundle adjustment algorithm and spatiotemporal loss function significantly improved triangulation accuracy.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2022)
Article
Computer Science, Information Systems
Renshu Gu, Zhongyu Jiang, Gaoang Wang, Kevin McQuade, Jenq-Neng Hwang
Summary: This paper proposes an unsupervised universal hierarchical 3D human pose estimation method that optimizes torso and limb poses, addressing the challenges of multi-person 3D pose estimation using a monocular freely moving camera.
MULTIMEDIA TOOLS AND APPLICATIONS
(2022)
Article
Biotechnology & Applied Microbiology
Shengyun Liang, Yu Zhang, Yanan Diao, Guanglin Li, Guoru Zhao
Summary: This study developed a markerless pose estimation system and used sample entropy to quantify the dynamic signal irregularity of gait. The validity and reliability of the system were assessed using various methods, and the agreement between markerless and marker-based measurements was evaluated using Bland-Altman analysis.
FRONTIERS IN BIOENGINEERING AND BIOTECHNOLOGY
(2022)
Review
Engineering, Multidisciplinary
Jianchu Lin, Shuang Li, Hong Qin, Hongchang Wang, Ning Cui, Qian Jiang, Haifang Jian, Gongming Wang
Summary: This article summarizes the research progress and methods related to 3D human pose estimation, focusing on the estimation of monocular RGB images and videos. It discusses common problems such as occlusion and poor model generalization, and introduces various approaches, including multi-view and weakly supervised learning methods. The article also emphasizes the issue of insufficient training datasets and discusses emerging datasets and evaluation indicators. Overall, it provides a useful overview of the field and can guide researchers, while also highlighting areas for further research.
CMES-COMPUTER MODELING IN ENGINEERING & SCIENCES
(2023)
Article
Computer Science, Artificial Intelligence
Junting Dong, Qi Fang, Wen Jiang, Yurou Yang, Qixing Huang, Hujun Bao, Xiaowei Zhou
Summary: This paper addresses the problem of reconstructing 3D poses of multiple people from a few calibrated camera views. The proposed approach uses a multi-way matching algorithm to cluster detected 2D poses and infer the 3D poses of each person efficiently. It also combines geometric and appearance cues for cross-view matching and proposes an efficient tracking method.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Computer Science, Artificial Intelligence
Shu Chen, Yaxin Xu, Zhengdong Pu, Jianquan Ouyang, Beiji Zou
Summary: SkeletonPose is a method that improves the accuracy of 3D human pose estimation by using human skeleton constraints. By combining data-driven and calculation methods, the proposed approach regresses the z-coordinate of the root joint using deep convolutional networks and calculates the 3D human pose based on the skeleton length invariance constraint. This method reduces pose estimation errors by considering the skeleton length prior. Evaluation results show that SkeletonPose achieves better performance compared to other state-of-the-art pose estimation approaches.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Ji Yang, Youdong Ma, Xinxin Zuo, Sen Wang, Minglun Gong, Li Cheng
Summary: This paper proposes a method that jointly tackles the tasks of estimating 3D human body poses and predicting future 3D motions. The method utilizes a self-projection mechanism, a multi-task architecture, and a global refinement module to improve performance. Experimental results show competitive performance compared to existing techniques.
PATTERN RECOGNITION
(2022)
Article
Multidisciplinary Sciences
Tobias Baumgartner, Benjamin Paassen, Stefanie Klatt
Summary: Collecting large datasets for investigations into human locomotion is expensive and labor-intensive. Accurate methods for 3D human pose estimation in the wild could assist with collecting datasets for analyzing running kinematics from TV broadcast data. However, current state-of-the-art 3D human pose estimation methods are not yet accurate enough for kinematics research, as small differences in 3D angles play a significant role in biomechanical research.
SCIENTIFIC REPORTS
(2023)
Article
Engineering, Biomedical
Nazim Haouchine, Parikshit Juvekar, Michael Nercessian, William Wells, Alexandra Golby, Sarah Frisken
Summary: This study presents a novel Augmented Reality neurosurgical system that superimposes pre-operative 3D meshes derived from MRI onto a view of the brain surface during surgery. The method uses cortical vessels as main features for rigid and non-rigid 3D/2D registration, and achieves low pose error.
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING
(2022)
Article
Computer Science, Artificial Intelligence
Aiden Nibali, Joshua Millward, Zhen He, Stuart Morgan
Summary: This paper introduces a method for obtaining 3D human pose annotations using only three video cameras, and constructs the ASPset-510 dataset. By using this dataset, improvements in pose model generalization on established benchmarks can be achieved.
IMAGE AND VISION COMPUTING
(2021)
Article
Computer Science, Artificial Intelligence
Zhongyang Zhang, Kaidong Chai, Haowen Yu, Ramzi Majaj, Francesca Walsh, Edward Wang, Upal Mahbub, Hava Siegelmann, Donghyun Kim, Tauhidur Rahman
Summary: Technology-mediated dance experiences are crucial in both traditional and virtual reality-based gaming platforms. Current solutions for dance gaming primarily use RGB or RGB-Depth cameras, which have limitations such as low-light conditions and excessive power consumption. Neuromorphic cameras, with their ultra-low latency and energy efficiency, provide a viable solution to overcome these limitations.
Article
Computer Science, Artificial Intelligence
Ruixu Liu, Ju Shen, He Wang, Chen Chen, Sen-ching Cheung, Vijayan K. Asari
Summary: The study presents a systematic approach for incorporating traditional networks and constraints into an attention framework for learning long-range dependencies and achieving end-to-end pose estimation. By adapting temporal receptive fields through a multi-scale dilated convolution structure, real-time performance is achieved with state-of-the-art results.
INTERNATIONAL JOURNAL OF COMPUTER VISION
(2021)
Article
Multidisciplinary Sciences
Alice Ruget, Max Tyler, German Mora Martin, Stirling Scholes, Feng Zhu, Istvan Gyongy, Brent Hearn, Steve McLaughlin, Abderrahim Halimi, Jonathan Leach
Summary: Single-photon-sensitive depth sensors have a growing application in human pose and gesture recognition in next-generation electronics. This study presents a temporal to spatial mapping to significantly enhance the resolution of a simple time-of-flight sensor. The developed explainable framework offers insight into how the network utilizes input data and relevant parameters.
Article
Multidisciplinary Sciences
Laurie Needham, Murray Evans, Darren P. Cosker, Logan Wade, Polly M. McGuigan, James L. Bilzon, Steffi L. Colyer
Summary: Markerless pose estimation algorithms show potential for large scale movement studies outside of the laboratory, but their accuracy is not yet fully evaluated. 3D joint centre locations derived from these algorithms are not consistently comparable to marker-based motion capture.
SCIENTIFIC REPORTS
(2021)