Kamery v dohledov\u00fdch syst\u00e9mech sn\u00edmaj\u00ed t\u011bla, ne tv\u00e1\u0159e. Pr\u00e1v\u011b ve chv\u00edl\u00edch, kdy je obli\u010dej zakryt\u00fd, rozmazan\u00fd nebo zachycen\u00fd z nevhodn\u00e9ho \u00fahlu, tradi\u010dn\u00ed biometrick\u00e9 metody selh\u00e1vaj\u00ed. Lze \u010dlov\u011bka spolehliv\u011b identifikovat jen podle proporc\u00ed jeho t\u011bla \u2014 bez tv\u00e1\u0159e, bez oble\u010den\u00ed, bez ch\u016fze?<\/p>\n\n\n\n
Tato pr\u00e1ce, prezentovan\u00e1 na konferenci IIT.SRC 2026, ukazuje, \u017ee to v omezen\u00e9 m\u00ed\u0159e mo\u017en\u00e9 je. Syst\u00e9m extrahuje 23 anatomicky v\u00fdznamn\u00fdch vzd\u00e1lenost\u00ed kloub\u016f z 3D modelu t\u011bla rekonstruovan\u00e9ho z b\u011b\u017en\u00e9 RGB kamery a pomoc\u00ed neuronov\u00e9 s\u00edt\u011b je p\u0159ev\u00e1d\u00ed do diskriminativn\u00edho prostoru p\u0159\u00edznak\u016f vhodn\u00e9ho pro identifikaci osob.<\/p>\n\n\n\n
Z monokul\u00e1rn\u00edch RGB sn\u00edmk\u016f je pomoc\u00ed metody Neural Localizer Fields (NLF) rekonstruov\u00e1n parametrick\u00fd 3D model t\u011bla ve form\u00e1tu SMPL. Z tohoto modelu je extrahov\u00e1no 22 normalizovan\u00fdch vzd\u00e1lenost\u00ed kloub\u016f tvo\u0159\u00edc\u00edch biometrick\u00fd podpis nez\u00e1visl\u00fd na vzd\u00e1lenosti od kamery. Surov\u00e9 p\u0159\u00edznaky vykazuj\u00ed vysokou variabilitu uvnit\u0159 t\u0159\u00edd, proto je navr\u017eena embeddingov\u00e1 neuronov\u00e1 s\u00ed\u0165 tr\u00e9novan\u00e1 pomoc\u00ed triplet a quadruplet loss funkc\u00ed, kter\u00e1 p\u0159\u00edznaky mapuje do hyperkulov\u00e9 diskriminativn\u00ed reprezentace. Spr\u00e1vn\u00e1 identita je pot\u00e9 vyhled\u00e1na pomoc\u00ed kosinov\u00e9 podobnosti.<\/p>\n\n\n\n
Experimenty prob\u011bhly na t\u0159ech ve\u0159ejn\u00fdch datasetech (HuMMan, AMASS, MVHuman) celkem s 242 identitami. Na \u010dist\u00fdch 3D datech dos\u00e1hl syst\u00e9m p\u0159ibli\u017en\u011b 95% p\u0159esnosti Top-10 i p\u0159i galerii 160 osob. P\u0159i rekonstrukci z 2D sn\u00edmk\u016f syst\u00e9m spr\u00e1vn\u011b za\u0159adil hledanou osobu mezi p\u011bt nejlep\u0161\u00edch kandid\u00e1t\u016f ve v\u00edce ne\u017e 90 % p\u0159\u00edpad\u016f.<\/p>\n\n\n\n
Je v\u0161ak nutn\u00e9 zasadit tyto v\u00fdsledky do spr\u00e1vn\u00e9ho kontextu. T\u011blesn\u00e9 proporce pat\u0159\u00ed mezi nejm\u00e9n\u011b spolehliv\u00e9 biometrick\u00e9 modality \u2014 na rozd\u00edl od otisku prstu, duhovky nebo \u017eil prstu jsou mezi lidmi mnohem m\u00e9n\u011b jedine\u010dn\u00e9 a sn\u00e1ze zam\u011bniteln\u00e9. Metoda proto nen\u00ed navr\u017eena jako prim\u00e1rn\u00ed identifika\u010dn\u00ed n\u00e1stroj, ale jako z\u00e1lo\u017en\u00ed modalita pro situace, kdy \u017e\u00e1dn\u00fd spolehliv\u011bj\u0161\u00ed identifik\u00e1tor k dispozici nen\u00ed.<\/p>\n\n\n\n
I p\u0159es inherentn\u00ed omezen\u00ed t\u011blesn\u00fdch proporc\u00ed jako biometrick\u00e9ho znaku metoda nab\u00edz\u00ed prakticky vyu\u017eiteln\u00fd v\u00fdsledek: bez speci\u00e1ln\u00edho hardware, invariantn\u011b v\u016f\u010di oble\u010den\u00ed i sv\u011bteln\u00fdm podm\u00ednk\u00e1m a s mo\u017enost\u00ed p\u0159id\u00e1vat nov\u00e9 osoby bez p\u0159etr\u00e9nov\u00e1n\u00ed modelu. T\u011blesn\u00e9 proporce tak p\u0159edstavuj\u00ed u\u017eite\u010dnou z\u00e1lo\u017en\u00ed biometrickou modalitu pro forenzn\u00ed a bezpe\u010dnostn\u00ed aplikace \u2014 v\u0161ude tam, kde prim\u00e1rn\u00ed identifik\u00e1tory nejsou k dispozici.<\/p>\n\n\n\n
Samuel \u0160im\u00fan, Tom\u00e1\u0161 Goldmann, Jan Pluskal\u00a0Exploring the Discriminative Power of SMPL-Derived Body Proportions: An Initial Investigation<\/em>\u00a0Podpo\u0159eno FIT VUT Brno, grant FIT-S-26-9011.<\/p>\n","protected":false},"excerpt":{"rendered":" Kamery v dohledov\u00fdch syst\u00e9mech sn\u00edmaj\u00ed t\u011bla, ne tv\u00e1\u0159e. Pr\u00e1v\u011b ve chv\u00edl\u00edch, kdy je obli\u010dej zakryt\u00fd, rozmazan\u00fd nebo zachycen\u00fd z nevhodn\u00e9ho \u00fahlu, tradi\u010dn\u00ed biometrick\u00e9 metody selh\u00e1vaj\u00ed. Lze \u010dlov\u011bka spolehliv\u011b identifikovat jen podle proporc\u00ed jeho t\u011bla \u2014 bez tv\u00e1\u0159e, bez oble\u010den\u00ed, bez ch\u016fze? Tato pr\u00e1ce, prezentovan\u00e1 na konferenci IIT.SRC 2026, ukazuje, \u017ee […]<\/p>\n","protected":false},"author":3,"featured_media":1341,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"ub_ctt_via":"","footnotes":""},"categories":[1],"tags":[],"class_list":["post-1338","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-strade"],"featured_image_src":"https:\/\/strade.fit.vutbr.cz\/wp-content\/uploads\/2026\/05\/smpl_detail.png","author_info":{"display_name":"igoldmann","author_link":"https:\/\/strade.fit.vutbr.cz\/index.php\/author\/igoldmann\/"},"_links":{"self":[{"href":"https:\/\/strade.fit.vutbr.cz\/index.php\/wp-json\/wp\/v2\/posts\/1338","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/strade.fit.vutbr.cz\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/strade.fit.vutbr.cz\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/strade.fit.vutbr.cz\/index.php\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/strade.fit.vutbr.cz\/index.php\/wp-json\/wp\/v2\/comments?post=1338"}],"version-history":[{"count":3,"href":"https:\/\/strade.fit.vutbr.cz\/index.php\/wp-json\/wp\/v2\/posts\/1338\/revisions"}],"predecessor-version":[{"id":1343,"href":"https:\/\/strade.fit.vutbr.cz\/index.php\/wp-json\/wp\/v2\/posts\/1338\/revisions\/1343"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/strade.fit.vutbr.cz\/index.php\/wp-json\/wp\/v2\/media\/1341"}],"wp:attachment":[{"href":"https:\/\/strade.fit.vutbr.cz\/index.php\/wp-json\/wp\/v2\/media?parent=1338"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/strade.fit.vutbr.cz\/index.php\/wp-json\/wp\/v2\/categories?post=1338"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/strade.fit.vutbr.cz\/index.php\/wp-json\/wp\/v2\/tags?post=1338"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}