Deep Learning-based Human Pose Estimation: A Survey

Abstract

Human pose estimation aims to locate the human body parts and build human body representation (e.g., body skeleton) from input data such as images and videos. It has drawn increasing attention during the past decade and has been utilized in a wide range of applications including human-computer interaction, motion analysis, augmented reality, and virtual reality. Although the recently developed deep learning-based solutions have achieved high performance in human pose estimation, there still remain challenges due to insufficient training data, depth ambiguities, and occlusion. The goal of this survey paper is to provide a comprehensive review of recent deep learning-based solutions for both 2D and 3D pose estimation via a systematic analysis and comparison of these solutions based on their input data and inference procedures. More than 260 research papers since 2014 are covered in this survey. Furthermore, 2D and 3D human pose estimation datasets and evaluation metrics are included. Quantitative performance comparisons of the reviewed methods on popular datasets are summarized and discussed. Finally, the challenges involved, applications, and future research directions are concluded. A regularly updated project page is provided: https://github.com/zczcwh/DL-HPE

Lightning Pose: improved animal pose estimation via semi-supervised learning, Bayesian ensembling, and cloud-native open-source tools

“…Human pose estimation, like animal pose estimation, is most commonly approached using supervised heatmap regression on a frame-by-frame basis [49]. Unlike the animal setting, human models are trained on much larger labeled datasets containing either annotated images [50] or 3D motion capture [51].…”

“…In contrast, most animal pose estimation must contend with relatively scarce labels, lower quality videos, and bespoke sets of labels to track, varying by species and lab. Though human pose estimation models can impressively track crowds of moving humans, doing downstream science using the keypoints still presents several challenges [49] similar to those discussed in the Results. Lightning Pose can be applied to single-human pose estimation, by fine-tuning an off-the-shelf human pose estimation backbone to specific experimental setups (such as patients in a clinic), while enforcing our spatiotemporal constraints (or new ones).…”

Unveiling the Landscape of Human Pose Estimation

“…The survey by Holte et al provides a comparative analysis of early works (before 2012) in human pose estimation and activity recognition from multi-view videos (Holte et al, 2012). A recent comprehensive survey explores Human Pose Estimation (HPE) by categorizing methods based on 2D or 3D scenarios, single-view or multi-view approaches, and diverse data sources, employing various learning paradigms C. Zheng (2023). However, this survey primarily focuses on reviewing different approaches and their accuracy results, without delving into common methodologies.…”

GYM Grace Tracker

“…Reimers et al (2012) [1] utilized open pose for pose recognition, while Muhammad Usama et al (2017) [4] employed Microsoft Kinect to obtain real-time human joint points for identifying yoga poses. Shih-En et al (2016) [5] proposed an architecture leveraging multiple convolutional networks to enhance joint estimates, and Bogo et al (2016) [7] estimated 3D pose and mesh shape from a single RGB image.…”