The accuracy of several pose estimation methods for 3D joint centre localisation

Laurie Needham,Steffi L Colyer,James L Bilzon,Logan Wade,Darren P Cosker,Murray Evans,Polly M Mcguigan

doi:10.1038/s41598-021-00212-x

Laurie Needham, Steffi L Colyer + Show 5 more

Open Access

https://doi.org/10.1038/s41598-021-00212-x

Copy DOI

Journal: Scientific Reports	Publication Date: Oct 19, 2021
Citations: 55	License type: open-access

Affiliation: University of Bath

Abstract

Human movement researchers are often restricted to laboratory environments and data capture techniques that are time and/or resource intensive. Markerless pose estimation algorithms show great potential to facilitate large scale movement studies ‘in the wild’, i.e., outside of the constraints imposed by marker-based motion capture. However, the accuracy of such algorithms has not yet been fully evaluated. We computed 3D joint centre locations using several pre-trained deep-learning based pose estimation methods (OpenPose, AlphaPose, DeepLabCut) and compared to marker-based motion capture. Participants performed walking, running and jumping activities while marker-based motion capture data and multi-camera high speed images (200 Hz) were captured. The pose estimation algorithms were applied to 2D image data and 3D joint centre locations were reconstructed. Pose estimation derived joint centres demonstrated systematic differences at the hip and knee (~ 30–50 mm), most likely due to mislabeling of ground truth data in the training datasets. Where systematic differences were lower, e.g., the ankle, differences of 1–15 mm were observed depending on the activity. Markerless motion capture represents a highly promising emerging technology that could free movement scientists from laboratory environments but 3D joint centre locations are not yet consistently comparable to marker-based motion capture.

Highlights

Human movement researchers are often restricted to laboratory environments and data capture techniques that are time and/or resource intensive
Automated markerbased motion capture systems are commonplace in laboratory environments providing marker tracking with sub-millimeter accuracy[18] and greatly reducing the processing time when compared to manual data annotation
Marker placements often do not correspond directly to the true anatomical joint centres they are representing[20], and soft tissue artefact can add further measurement error[21,22]. These limitations of marker-based motion capture systems have been well studied in relation to “gold standard” methods such as bi-planar videoradiography, where marker-based errors up to 30 mm have been reported for lower limb joints centre locations[21,22]

Summary

Introduction

Human movement researchers are often restricted to laboratory environments and data capture techniques that are time and/or resource intensive. Joint centres such as the hip, knee or ankle, are estimated in order to reason the position and orientation of the body This process represents a challenging problem as the algorithm should be invariant to changes in scale, perspective, lighting and even partial occlusion of a body part. A forward pass of an image returns a 2D confidence map of key point locations (e.g., hip or elbow joint centres) in relation to that image alone The accuracy of these CNN based methods is typically evaluated against hand-labelled ground truth data which are undoubtably subject to human e rror[10,17] and not necessarily a true gold standard measure

Objectives

Methods

Discussion

Conclusion