As head of the Bio­MovE research pole at UTC’s Bio­me­chan­ics and Bio-engi­neer­ing Lab­o­ra­to­ry (BMBI), Khalil Ben Man­sour has been work­ing on the MoCapia project for two years now. Since its launch, the project has suc­ces­sive­ly involved six UTC interns. Yiyang Huang and Macéo Nar­bon, 3rd UTC stu­dents cur­rent­ly major­ing in Com­put­er Sci­ences and Engi­neer­ing stu­dents are involved here.

What are the project’s objec­tives? “Our research aims to devel­op solu­tions and bio­me­chan­i­cal mod­els that will, in par­tic­u­lar, allow us to analyse a giv­en person’s health sta­tus, assess their per­for­mance, and pre­vent mus­cu­loskele­tal dis­or­ders,” he explains.

This research has appli­ca­tions in many fields, such as sports, health and ergonom­ics. Until now, one of the most com­mon­ly-used mea­sure­ment tech­niques has been 3D motion cap­ture. “In this case, we use a sys­tem of opto­elec­tron­ic cam­eras to film the move­ment. But this tech­nique requires a high lev­el of exper­tise, sig­nif­i­cant prepa­ra­tion time, and lengthy pro­cess­ing times. This makes it dif­fi­cult to use in a clin­i­cal set­ting, for exam­ple, and ulti­mate­ly lim­its its appli­ca­tion to lab­o­ra­to­ry research,” says Khalil Ben Mansour.

How­ev­er, with the rise in com­put­ing pow­er and the advent of AI-based pose esti­ma­tion mod­els, things are start­ing to change. espe­cial­ly since today, with sim­ple, com­mer­cial­ly avail­able cam­eras, it is pos­si­ble to mea­sure move­ments with­out hav­ing to attach mark­ers to the sub­ject. This will sig­nif­i­cant­ly reduce both the cost of equip­ment and the prepa­ra­tion and pro­cess­ing time.

Hence the idea behind the MoCapia project. “We quick­ly became inter­est­ed in this very recent tech­nol­o­gy, which has not yet been val­i­dat­ed from a sci­en­tif­ic stand­point, par­tic­u­lar­ly in terms of accu­ra­cy. We decid­ed to use arti­fi­cial intel­li­gence to devel­op a motion cap­ture tool that is easy to han­dle and intu­itive to use. Cur­rent­ly, we are work­ing to fine-tune and improve its accu­ra­cy. To do this, we are test­ing var­i­ous solu­tions to obtain com­pre­hen­sive and pre­cise data that

Beyond improv­ing the accu­ra­cy of this new tech­nol­o­gy, anoth­er idea has tak­en root with­in the team. “We decid­ed to use pose esti­ma­tors gen­er­at­ed by arti­fi­cial intel­li­gence. How­ev­er, exist­ing pose esti­ma­tors are based on a sin­gle cam­era because their pur­pose is not for clin­i­cal use, for exam­ple, but pri­mar­i­ly for video ani­ma­tion. Indeed, in a video game, what mat­ters is the flu­id­i­ty of move­ment. This is some­what “lim­it­ed” giv­en the com­plex­i­ty of con­stituent ele­ments of a giv­en move­ment that we would like to quan­ti­fy and analyse in more rig­or­ous fields of appli­ca­tion, such as med­i­cine or high-lev­el sports,” Khalil explains.

It was then that the team came up with the idea of cou­pling mul­ti­ple cam­eras to per­form a 3D recon­struc­tion of the move­ment. This work offers a num­ber of chal­lenges, such as cam­era cal­i­bra­tion and syn­chro­niza­tion, among oth­ers. “We’re work­ing con­stant­ly to improve the mod­el by increas­ing the num­ber of cam­eras to gath­er more data and by test­ing dif­fer­ent cal­i­bra­tion tech­niques. Ulti­mate­ly, the goal is to find the opti­mal com­bi­na­tions that allow us to get clos­er to our ref­er­ence mod­el and achieve a lev­el of pre­ci­sion com­pat­i­ble with appli­ca­tions in bio-med­i­cine or elite sports. If we take the med­ical sec­tor as an exam­ple, it is imper­a­tive for the pros­thetist to know the exact angle need­ed to accu­rate­ly fit a pros­the­sis,” explains Khalil Ben Mansour.

In addi­tion to the sci­en­tif­ic val­i­da­tion of this new tech­nol­o­gy, the MoCapia project pri­mar­i­ly aims to devel­op a graph­i­cal inter­face so that anyone—from phys­i­cal ther­a­pists to doc­tors, ergono­mists, or sports coaches—can use it eas­i­ly and intu­itive­ly. This is the chal­lenge that the 6 UTC interns have tak­en on in suc­ces­sion since the project began. At present, 3rd year stu­dents Yiyang Huang and Macéo Nar­bon­net have tak­en over from the four pre­vi­ous interns.

What are their respec­tive roles? “In addi­tion to improv­ing cam­era cal­i­bra­tion, I enhanced the 3D mod­el of the human body to ensure accu­rate esti­ma­tion of inter­nal and exter­nal rota­tions of the body seg­ments,” says Macéo Nar­bon­net. Yiyang Huang, for her part, has focused pri­mar­i­ly on usabil­i­ty. “I opti­mized the ergonom­ics of the graph­i­cal inter­face by adopt­ing an archi­tec­ture that allows mul­ti­ple tasks to be exe­cut­ed simul­ta­ne­ous­ly with­out bugs, mak­ing the soft­ware intu­itive and flu­id for the user,” she says.

MSD