Can Markerless Pose Estimation Algorithms Estimate 3D Mass Centre Positions and Velocities during Linear Sprinting Activities?

Laurie Needham,Murray Evans,Darren P. Cosker,Steffi L. Colyer

doi:10.3390/s21082889

Abstract

The ability to accurately and non-invasively measure 3D mass centre positions and their derivatives can provide rich insight into the physical demands of sports training and competition. This study examines a method for non-invasively measuring mass centre velocities using markerless human pose estimation and Kalman smoothing. Marker (Qualysis) and markerless (OpenPose) motion capture data were captured synchronously for sprinting and skeleton push starts. Mass centre positions and velocities derived from raw markerless pose estimation data contained large errors for both sprinting and skeleton pushing (mean ± SD = 0.127 ± 0.943 and −0.197 ± 1.549 m·s−1, respectively). Signal processing methods such as Kalman smoothing substantially reduced the mean error (±SD) in horizontal mass centre velocities (0.041 ± 0.257 m·s−1) during sprinting but the precision remained poor. Applying pose estimation to activities which exhibit unusual body poses (e.g., skeleton pushing) appears to elicit more erroneous results due to poor performance of the pose estimation algorithm. Researchers and practitioners should apply these methods with caution to activities beyond sprinting as pose estimation algorithms may not generalise well to the activity of interest. Retraining the model using activity specific data to produce more specialised networks is therefore recommended.

Highlights

The accurate measurement and assessment of athletes’ movement profiles during training and competition can provide coaches with insight into the physical demands of competition and permit the monitoring of an athlete’s physical capabilities over both acute and longitudinal time scales [1]
In winter Olympic sports such as skeleton and bobsleigh, sprinting ability and the ability to load the sled with a high velocity has been associated with high performance outcomes [2,3,4,5] and represents information that coaches can use for talent identification and athlete monitoring
The aim of this study was to evaluate the ability of convolutional neural network (CNN)-based pose-estimation (OpenPose) to estimate centre of mass (CoM) velocities during two sprinting activities and examine whether advanced filtering methods can enhance the performance of the vision-based athlete tracking system

Summary

Introduction

The accurate measurement and assessment of athletes’ movement profiles during training and competition can provide coaches with insight into the physical demands of competition and permit the monitoring of an athlete’s physical capabilities over both acute and longitudinal time scales [1]. The current gold standard in determining the athlete’s centre of mass (CoM) location during sprinting is through the use of marker-based motion capture [6] Such systems can reconstruct the location of reflective markers with sub-millimetre accuracy [7] and when placed on the body in anatomically meaningful locations can be used to model the CoM locations of the segments (and thereafter the whole body) with a high precision [8]. The placement of markers is time consuming and may alter technique [9] To address these problems and allow for field-based collection of CoM displacement and velocity information, a range of technologies have emerged including manually annotated video analysis [10,11], laser distance measurement [12] and global or local positioning systems (GPS and LPS) technology [13]. FOFipgigeuunrrPeFOdeoAiepgsApe4euin4.crkt.PeOetOoAhypspep4ee.eo3nkOnDiePnPyportpoeeslonsceoieoPcnknoatkesstleteyioroycpuknpacoetsotyiieoniwpdnntoitsjidtonwhdietnetitdttethheeccectettehitncocieitotrnricnceoislenreecxfxleofiaexarmlfalmtimhlcplpeopclellolleoeselfsoustdurdd(ergereprreppeeipicpieccrtnrteitie)ninssnaeggegnnnppdttpoiionrnossigeggseehtethetsheset(eistrmliteelmidefmatf)ta(tiasgot(iitrndgoieoernedensnuedo)rdnufainu)ntrhgraidnienpnrgbdiuggosphrdphiutgyiun(shrsphgethr.idon(iB)nrjgelesguc.idd.teBe)eBcdlssiuliruoodcefneleetctshcsoirieodtrchfbceleeloptehsdi2sceyDdtd.betCiehpompuediacbi2ycgteD.testChthueeb22eDsD dOeppiecntpPtlhoaensee3.kDTehyriepscoeoxinnamsttlrpoulcecatdteieodmnjoosniwnsttirtachteetnshtterhecasitrflcoimlrebtfhislewl clieotcflhtoi(ungrgrreheeapnsr)oeacscneundrtrrienidgghfotthr(etrheleedfl)tesg(gisdrbeeusetnot)hfaitshnhedabsriolgadhrygte(plryreobdje)eecsntidecdeosroroencftttoehdtehdbeuor2diDnyg.imCuabgees pdlaenpeic.ttThtehh3iesD3efDxuasrmieocnpolpenrsdotcereumscsot. enHdsotjwroaeitnvetesrct,ehfnuatrtrthleisemrfbiosrssuwtheisetcclahenfintbg(egehrneasesenoe)ncacwnuidrtrhertidghehfotler(frttheaderm)leswgidshebesrueotfOtthphieesnhPbaoossdelyahrpagsreodljeyetcebtceeteeddnojconoitnrort etchcetneetd2reDsduimriangge thpthelae3n3DeDw. fTifutuhssiilsoiiomnenxiptapermdroocpsceulesecsscsd.e.sHeHsmoionwownmeesvutverletarir,pt,eflufesurftrtihehtahledterslriiosmifsssuvbuieeseswwsc.icatancnhbibneeegennhsaseseeeonncwcwuitrihtrhetdhthefeolrelefttfhtaearlrmemgwswbhhueertretehOOisppheenanPsPoloasrsegehehalyassbddeeeetetneccctetoedrdrjeojcoitnientdtccdeenuntrrtirenesgs wwitihthlilmimitieteddsusucccceessssininmmuultlitpiplelefifeiledldssoof fvvieiwew. .

Objectives

Methods

Results

Discussion

Conclusion