{"id":48685,"date":"2025-08-18T17:52:04","date_gmt":"2025-08-18T15:52:04","guid":{"rendered":"https:\/\/www.derivaty.sk\/?p=48685"},"modified":"2025-08-18T17:52:04","modified_gmt":"2025-08-18T15:52:04","slug":"detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii","status":"publish","type":"post","link":"https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/","title":{"rendered":"Detekcia a vyh\u00fdbanie (DAA): Vyu\u017eitie hlbok\u00e9ho u\u010denia pre prevenciu kol\u00edzi\u00ed"},"content":{"rendered":"<h2>Detekcia a vyh\u00fdbanie<\/h2>\n<p>Detekcia a vyh\u00fdbanie (<em>Detect and Avoid, DAA<\/em>) je k\u013e\u00fa\u010dov\u00e1 schopnos\u0165 auton\u00f3mnych bezpilotn\u00fdch lietadiel (UAS), ktor\u00e1 umo\u017e\u0148uje v\u010das rozpozna\u0165 prek\u00e1\u017eky a kooperat\u00edvne aj nekooperat\u00edvne \u00fa\u010dastn\u00edky vzdu\u0161n\u00e9ho priestoru a bezpe\u010dne upravi\u0165 trajekt\u00f3riu. V posledn\u00fdch rokoch do\u0161lo k z\u00e1sadn\u00e9mu posunu od ru\u010dne navrhnut\u00fdch pravidiel k d\u00e1tovo riaden\u00fdm pr\u00edstupom: u\u010deniu s u\u010dite\u013eom (supervised) a bez u\u010dite\u013ea (unsupervised), pr\u00edpadne k ich hybridom (semi-supervised, self-supervised). Cie\u013eom \u010dl\u00e1nku je systematicky pop\u00edsa\u0165 architekt\u00fary, senzorick\u00e9 modality, tr\u00e9novacie a valida\u010dn\u00e9 procesy, metriky, aj nasadzovanie DAA na palubn\u00fdch v\u00fdpo\u010dtov\u00fdch platform\u00e1ch s oh\u013eadom na bezpe\u010dnos\u0165, latenciu a regul\u00e1cie.<\/p>\n<h2>Probl\u00e9mov\u00e1 formul\u00e1cia DAA<\/h2>\n<p>DAA m\u00f4\u017eeme formulova\u0165 ako kombin\u00e1ciu \u00faloh po\u010d\u00edta\u010dov\u00e9ho videnia a riadenia: (1) <em>percepcia<\/em> (detekcia objektov, odhad vzdialenosti a relat\u00edvnej r\u00fdchlosti, sledovanie cie\u013eov), (2) <em>predikcia<\/em> (odhad bud\u00facich trajekt\u00f3ri\u00ed), (3) <em>rozhodovanie<\/em> (vo\u013eba vyh\u00fdbacej man\u00e9vre v s\u00falade s obmedzeniami dynamiky a pravidlami vzdu\u0161n\u00e9ho priestoru). Pri u\u010den\u00ed s u\u010dite\u013eom sa modely u\u010dia priamo z ozna\u010den\u00fdch d\u00e1t (detek\u010dn\u00e9 r\u00e1miky, triedy, vzdialenosti), pri u\u010den\u00ed bez u\u010dite\u013ea sa vyu\u017e\u00edvaj\u00fa latentn\u00e9 reprezent\u00e1cie, anom\u00e1lie, alebo konzistencia medzi senzormi a \u010dasom, aby sa <em>vydolovali<\/em> sign\u00e1ly bez nutnosti rozsiahleho ru\u010dn\u00e9ho ozna\u010dovania.<\/p>\n<h2>Senzorick\u00e9 modality a f\u00fazia<\/h2>\n<ul>\n<li><strong>RGB\/IR kamery<\/strong>: vysok\u00e1 rozli\u0161ovacia schopnos\u0165, n\u00edzka hmotnos\u0165; citlivos\u0165 na osvetlenie a po\u010dasie. IR roz\u0161iruje pou\u017eite\u013enos\u0165 v noci.<\/li>\n<li><strong>Stereo\/\u0161\u00edrkov\u00e9 kamery<\/strong>: hust\u00e1 parallax inform\u00e1cia, okam\u017eit\u00fd odhad h\u013abky na kr\u00e1tke a stredn\u00e9 vzdialenosti.<\/li>\n<li><strong>Lidar<\/strong>: presn\u00e1 3D geometria, robustn\u00e1 vo\u010di osvetleniu; hmotnos\u0165, spotreba a cena s\u00fa kompromisy.<\/li>\n<li><strong>Radar\/FMCW<\/strong>: dlh\u00fd dosah a priamo odhadnut\u00e1 radi\u00e1lna r\u00fdchlos\u0165, dobr\u00e1 priechodnos\u0165 nepriazniv\u00fdm po\u010das\u00edm.<\/li>\n<li><strong>ADS-B\/kooperat\u00edvne zdroje<\/strong>: identifik\u00e1cia a poloha kooperat\u00edvnych lietadiel; nepokr\u00fdva mal\u00e9\/nekoperat\u00edvne objekty (vt\u00e1ky, drony).<\/li>\n<li><strong>IMU\/GNSS<\/strong>: vlastn\u00fd stav nosi\u010da pre relat\u00edvne odhady a stabiliz\u00e1ciu vizu\u00e1lnych met\u00f3d.<\/li>\n<\/ul>\n<p>F\u00fazia m\u00f4\u017ee by\u0165 <em>early<\/em> (spojenie surov\u00fdch d\u00e1t), <em>mid-level<\/em> (spojenie hlavn\u00fdch pr\u00edznakov) alebo <em>late<\/em> (spojenie rozhodnut\u00ed). Pri u\u010den\u00fdch modeloch je efekt\u00edvna mid-level f\u00fazia s priestorovo-\u010dasov\u00fdmi modulmi (3D konvol\u00facie, attention nad sekvenciami, transformerov\u00e9 enk\u00f3dery) a s extenziou o fyzik\u00e1lne obmedzenia (kone\u010dn\u00e1 dynamika vlastn\u00e9ho UAV).<\/p>\n<h2>U\u010denie s u\u010dite\u013eom: detekcia, segment\u00e1cia a metrikuje h\u013abky<\/h2>\n<p>Pri u\u010den\u00ed s u\u010dite\u013eom s\u00fa k\u013e\u00fa\u010dov\u00e9 kvalitn\u00e9 anot\u00e1cie a reprezentat\u00edvnos\u0165 scen\u00e1rov. Typick\u00e9 podprobl\u00e9my:<\/p>\n<ul>\n<li><strong>Detekcia objektov<\/strong>: modely jednorazov\u00e9ho priechodu (anchor-free detektory) poskytuj\u00fa n\u00edzku latenciu pre mal\u00e9 objekty v dia\u013eke. D\u00f4le\u017eit\u00e1 je tr\u00e9ningov\u00e1 strat\u00e9gia s <em>focal loss<\/em> a \u0165a\u017ebou \u0165a\u017ek\u00fdch negat\u00edvov, ke\u010f\u017ee triedna nerovnov\u00e1ha je extr\u00e9mna.<\/li>\n<li><strong>In\u0161tan\u010dn\u00e1\/semantick\u00e1 segment\u00e1cia<\/strong>: presn\u00e9 obrysy pre prek\u00e1\u017eky (k\u00e1ble, stromy) umo\u017e\u0148uj\u00fa jemn\u00e9 vyh\u00fdbacie man\u00e9vre. Kombin\u00e1cie encoder-decoder architekt\u00far s priestorovo-\u010dasovou rekurenciou zni\u017euj\u00fa chvenie masiek.<\/li>\n<li><strong>Monokul\u00e1rna\/stereo h\u013abka<\/strong>: pri monokul\u00e1rnej h\u013abke sa u\u010d\u00ed regresia vzdialenosti; pri stereo je k dispoz\u00edcii doh\u013ead z disparity. Pre DAA je kritick\u00e1 <em>metricky \u0161k\u00e1lovan\u00e1<\/em> h\u013abka a spo\u013eahlivostn\u00e9 mapy (uncertainty).<\/li>\n<li><strong>Viackan\u00e1lov\u00e1 f\u00fazia (kamera+radar\/lidar)<\/strong>: pri tr\u00e9novan\u00ed sa vyu\u017e\u00edva <em>teacher-student<\/em> paradigmu (lidar ako u\u010dite\u013e, kamera ako \u0161tudent) na prenos geometrie do lacnej\u0161ej modality.<\/li>\n<\/ul>\n<h2>U\u010denie bez u\u010dite\u013ea: anom\u00e1lie, konzistencia a self-supervised<\/h2>\n<p>U\u010denie bez u\u010dite\u013ea zmier\u0148uje z\u00e1vislos\u0165 od anot\u00e1ci\u00ed a zlep\u0161uje generaliz\u00e1ciu:<\/p>\n<ul>\n<li><strong>Detekcia anom\u00e1li\u00ed<\/strong>: autoenkod\u00e9ry a <em>normalizing flows<\/em> sa u\u010dia model norm\u00e1lnych vizu\u00e1lnych stavov; odch\u00fdlky (prete\u010denie rekon\u0161truk\u010dnej chyby) indikuj\u00fa nezn\u00e1me prek\u00e1\u017eky alebo ne\u010dakan\u00e9 situ\u00e1cie.<\/li>\n<li><strong>Kontrast\u00edvne u\u010denie<\/strong>: vytv\u00e1ra reprezent\u00e1cie invariantn\u00e9 vo\u010di osvetleniu, \u0161umu, perspekt\u00edve; zni\u017euje potrebu anot\u00e1ci\u00ed a zlep\u0161uje prenos do nov\u00fdch podmienok.<\/li>\n<li><strong>Self-supervised h\u013abka a ego-motion<\/strong>: uklad\u00e1 sa fotometrick\u00e1 konzistencia medzi po sebe id\u00facimi sn\u00edmkami a maskami pohybu; v\u00fdstupom je hust\u00e1 h\u013abka a odhad pohybu bez GT h\u013abky.<\/li>\n<li><strong>Clustering a pseudo-labels<\/strong>: neozna\u010den\u00e9 d\u00e1ta sa zhlukuj\u00fa v priestore pr\u00edznakov a n\u00e1sledne sa iterat\u00edvne dotv\u00e1raj\u00fa pseudo-\u0161t\u00edtky pre \u010fal\u0161\u00ed supervised tr\u00e9ning.<\/li>\n<\/ul>\n<h2>Sledovanie a odhad relat\u00edvneho stavu<\/h2>\n<p>Percepciu dop\u013a\u0148a multiobjektov\u00e9 sledovanie (MOT) a odhad relat\u00edvnej polohy a r\u00fdchlosti cie\u013ea. Kombinuje sa u\u010den\u00fd detektor s filtrami (Kalman\/IMM\/UKF) a <em>tracking-by-detection<\/em> paradigmou. Pre r\u00fdchle ciele s\u00fa vhodn\u00e9 asocia\u010dn\u00e9 met\u00f3dy s <em>appearance embeddings<\/em> (reidentifik\u00e1cia) a s konzistenciou optick\u00e9ho toku. Pre senzory s priamym odhadom r\u00fdchlosti (radar) sa vyu\u017e\u00edva pravdepodobnostn\u00e1 f\u00fazia (Bayes, factor graphs) s kamerou.<\/p>\n<h2>Predikcia trajekt\u00f3ri\u00ed a interak\u010dn\u00e9 modely<\/h2>\n<p>Predikcia bud\u00facich stavov objektov zoh\u013ead\u0148uje interakcie a obmedzenia man\u00e9vrovate\u013enosti. U\u010den\u00e9 <em>graph neural networks<\/em> a transformerov\u00e9 sekven\u010dn\u00e9 modely dok\u00e1\u017eu extrapolova\u0165 kooperat\u00edvne aj nekooperat\u00edvne spr\u00e1vanie. V praxi sa via\u017eu fyzik\u00e1lne limity (max. zr\u00fdchlenie\/ot\u00e1\u010danie) a \u201esoci\u00e1lne\u201c pravidl\u00e1 (bezpe\u010dn\u00e9 odstupy) formou <em>loss<\/em> penaliz\u00e1ci\u00ed alebo ako <em>hard constraints<\/em> na v\u00fdstupe.<\/p>\n<h2>Rozhodovanie a pl\u00e1novanie vyh\u00fdbacej man\u00e9vre<\/h2>\n<p>V\u00fdstup percepcie a predikcie vstupuje do pl\u00e1nova\u010da. Pre bezpe\u010dnos\u0165 sa typicky vol\u00ed dvojstup\u0148ov\u00fd pr\u00edstup: (1) r\u00fdchla reakt\u00edvna vrstva (<em>velocity obstacles<\/em>, <em>dynamic window<\/em>), (2) vy\u0161\u0161ia optimaliza\u010dn\u00e1 vrstva (MPC, sampling-based planners), ktor\u00e1 re\u0161pektuje dynamiku a pravidl\u00e1. U\u010den\u00e9 politiky (napr. imit\u00e1cia expertov) m\u00f4\u017eu generova\u0165 inicializ\u00e1cie alebo priestor kandid\u00e1tov, zatia\u013e \u010do fin\u00e1lnu vo\u013ebu validuje bezpe\u010dnostn\u00fd <em>shield<\/em> s garantovan\u00fdmi invarantami (napr. <em>Control Barrier Functions<\/em>).<\/p>\n<h2>Neistota, kalibr\u00e1cia a <em>runtime assurance<\/em><\/h2>\n<p>Kritick\u00e1 nie je len presnos\u0165, ale aj spr\u00e1vna kalibr\u00e1cia pravdepodobnost\u00ed. Pou\u017e\u00edva sa <em>Monte Carlo dropout<\/em>, ensembling, eviden\u010dn\u00e9 hlbok\u00e9 u\u010denie a <em>temperature scaling<\/em> na \u00farove\u0148 d\u00f4very. V prev\u00e1dzke <em>runtime assurance<\/em> (RTA) monitoruje, \u010di riziko kol\u00edzie prekro\u010dilo prah alebo \u010di sa syst\u00e9m nach\u00e1dza mimo tr\u00e9ningovej distrib\u00facie (OOD). Pri prekro\u010den\u00ed limitov sa aktivuje konzervat\u00edvna politika (spomalenie, st\u00fapanie, n\u00fadzov\u00e9 zastavenie) alebo prechod na pravidlov\u00e9 z\u00e1lohy.<\/p>\n<h2>D\u00e1ta, syntetika a dom\u00e9nov\u00e1 adapt\u00e1cia<\/h2>\n<p>Re\u00e1lne kol\u00edzne situ\u00e1cie s\u00fa zriedkav\u00e9, preto je nutn\u00e1 syntetika a <em>domain randomization<\/em> (po\u010dasie, text\u00fary, osvetlenie, pohybov\u00e9 profily). Prechod do reality rie\u0161i <em>unsupervised domain adaptation<\/em> (adverzari\u00e1lne prisp\u00f4sobenie pr\u00edznakov, \u0161t\u00fdlov\u00fd transfer) a <em>test-time adaptation<\/em> (jemn\u00e9 doladenie \u0161tatist\u00edk BN). K\u013e\u00fa\u010dov\u00e9 je dop\u013a\u0148a\u0165 syntetiku re\u00e1lnymi <em>hard case<\/em> datasetmi: mal\u00e9 drony na pozad\u00ed oblohy, vt\u00e1ky, k\u00e1ble, zrkadlenia, protisvetlo, d\u00e1\u017e\u010f, hmla.<\/p>\n<h2>Metriky a hodnotenie v\u00fdkonu DAA<\/h2>\n<ul>\n<li><strong>Percepcia<\/strong>: mAP\/mAR pre mal\u00e9 objekty (\u0161pecifick\u00e9 pre vzdialen\u00e9 pixely), <em>Average Localization Error<\/em>, presnos\u0165 h\u013abky (RMSE, \u03b4-metriky), MOT (MOTA\/MOTP\/HOTA).<\/li>\n<li><strong>Predikcia<\/strong>: ADE\/FDE (priemern\u00e9 a koncov\u00e9 chyby trajekt\u00f3rie), pravdepodobnostn\u00e9 metriky (NLL, CRPS) a kalibr\u00e1cia predik\u010dn\u00fdch ku\u017ee\u013eov.<\/li>\n<li><strong>Bezpe\u010dnos\u0165<\/strong>: <em>Minimum Miss Distance<\/em> (MMD), <em>Time to Closest Point of Approach<\/em> (TCPA), <em>Probability of Collision<\/em> (PoC) pod spo\u013eahlivos\u0165ou senzoriky.<\/li>\n<li><strong>Prev\u00e1dzkov\u00e9<\/strong>: latencia end-to-end (kamera\u2192man\u00e9ver), vyu\u017eitie energie, dostupnos\u0165 (>99.9%), miera plan-abort vs. falo\u0161n\u00e9 poplachy (FAR).<\/li>\n<\/ul>\n<h2>Realtime implement\u00e1cia a optimaliz\u00e1cia<\/h2>\n<p>Palubn\u00e9 platformy (SoC s GPU\/NPU) vy\u017eaduj\u00fa optimaliz\u00e1cie: kvantiz\u00e1cia, prerez\u00e1vanie (pruning), <em>knowledge distillation<\/em>, f\u00fazia oper\u00e1ci\u00ed a pipelining (s\u00fabe\u017en\u00e9 z\u00edskavanie sn\u00edmkov, inferencia a pl\u00e1novanie). Kritick\u00e1 je deterministick\u00e1 latencia: fixn\u00e9 ve\u013ekosti d\u00e1vok, predalokovan\u00e1 pam\u00e4\u0165, rozdelenie jadier pre ROS2 exek\u00fator, prahovanie FPS pod\u013ea r\u00fdchlosti letu a rizika. Pre v\u00fdpadky senzorov sa udr\u017eiava minim\u00e1lne funk\u010dn\u00fd profil (napr. vizu\u00e1lny tok + IMU) s degradovan\u00fdm, ale bezpe\u010dn\u00fdm spr\u00e1van\u00edm.<\/p>\n<h2>Bezpe\u010dnostn\u00e9 argumenty a overovanie<\/h2>\n<p>DAA s u\u010den\u00fdmi modelmi mus\u00ed by\u0165 doplnen\u00e9 form\u00e1lnymi a \u0161trukt\u00farovan\u00fdmi d\u00f4kazmi bezpe\u010dnosti: <em>Requirements-to-Evidence<\/em> trasovanie, nez\u00e1visl\u00e9 testy, oddelen\u00e9 datasety (train\/val\/test s \u010dasoprostorovou segreg\u00e1ciou), <em>adversarial<\/em> testy (oslnenie, \u0161um, z\u00e1mern\u00e9 maskovanie), HIL\/SIL kampane so simul\u00e1ciou kol\u00edzi\u00ed a bl\u00edzkych preletov. <em>Runtime monitors<\/em> s <em>simplex<\/em> architekt\u00farou nech\u00e1vaj\u00fa u\u010den\u00e9mu modulu generova\u0165 n\u00e1vrhy, ktor\u00e9 s\u00fa validovan\u00e9 pravidlov\u00fdm bezpe\u010dnostn\u00fdm jadrom.<\/p>\n<h2>Supervised vs. unsupervised: komparat\u00edvne v\u00fdhody<\/h2>\n<ul>\n<li><strong>U\u010denie s u\u010dite\u013eom<\/strong>: vysok\u00e1 presnos\u0165 pre zn\u00e1me triedy a situ\u00e1cie, predv\u00eddate\u013en\u00fd v\u00fdkon; n\u00e1ro\u010dn\u00e9 na anot\u00e1cie, riziko <em>dataset bias<\/em>.<\/li>\n<li><strong>U\u010denie bez u\u010dite\u013ea<\/strong>: odha\u013euje neo\u010dak\u00e1van\u00e9 objekty a stavy, \u0161k\u00e1luje s ve\u013ek\u00fdmi neozna\u010den\u00fdmi datasetmi; potrebuje opatrn\u00fa interpret\u00e1ciu (anom\u00e1lia \u2260 hrozba) a dobr\u00fa integr\u00e1ciu s rozhodovan\u00edm.<\/li>\n<\/ul>\n<p>V praxi v\u00ed\u0165az\u00ed <strong>hybrid<\/strong>: supervised percepcia pre zn\u00e1me hrozby (lietadl\u00e1, helikopt\u00e9ry, drony), unsupervised OOD detekcia a anom\u00e1lie ako <em>watchdog<\/em>, semi-supervised jemn\u00e9 doladenie na nov\u00fdch oblastiach nasadenia.<\/p>\n<h2>Integr\u00e1cia s pravidlami vzdu\u0161n\u00e9ho priestoru a auton\u00f3miou<\/h2>\n<p>DAA rozhodovanie mus\u00ed re\u0161pektova\u0165 koridory, no-fly z\u00f3ny, minim\u00e1lne v\u00fd\u0161ky, priority a pravidl\u00e1 <em>right-of-way<\/em>. Percep\u010dno-pl\u00e1novacia slu\u010dka m\u00e1 rozhranie k vy\u0161\u0161iemu pl\u00e1nova\u010du misie (trajekt\u00f3rie, body z\u00e1ujmu) a ku komunika\u010dnej vrstve pre kooper\u00e1ciu v rojov\u00fdch scen\u00e1roch (zdie\u013eanie zisten\u00ed, distribuovan\u00e1 f\u00fazia, vyh\u00fdbanie sa kol\u00edzi\u00e1m medzi \u010dlenmi roju).<\/p>\n<h2>Kooperat\u00edvne DAA v rojoch<\/h2>\n<p>V rojov\u00fdch oper\u00e1ci\u00e1ch DAA z\u00edskava syst\u00e9mov\u00fd rozmer: zdie\u013eanie lok\u00e1lnych m\u00e1p obsadenia, odhady trajekt\u00f3ri\u00ed a koordinovan\u00e9 man\u00e9vre. U\u010den\u00e9 <em>graph-based<\/em> modely s komunikuj\u00facimi uzlami (message passing) dok\u00e1\u017eu \u0161kalova\u0165, ak sa obmedz\u00ed \u0161\u00edrka p\u00e1sma cez kompresiu pr\u00edznakov a prioritiz\u00e1ciu hrozieb. Konflikty rie\u0161i arbitr\u00e1\u017e s glob\u00e1lnymi pravidlami a lok\u00e1lnymi <em>safety bubbles<\/em>.<\/p>\n<h2>Okrajov\u00e9 pr\u00edpady, robustnos\u0165 a <em>fail-operational<\/em><\/h2>\n<p>Medzi \u0165a\u017ek\u00e9 pr\u00edpady patria mal\u00e9, r\u00fdchle ciele proti preexponovan\u00e9mu nebu, drifty kamery pri vibr\u00e1ci\u00e1ch, d\u00e1\u017e\u010f\/hmla, reflexie na skle, ve\u013emi tenk\u00e9 prek\u00e1\u017eky (lan\u00e1). Robustnos\u0165 sa zvy\u0161uje d\u00e1tovou augment\u00e1ciou (simulovan\u00e9 oslnenie, kvapky, mot\u00fd\u013eov\u00fd \u0161um), fyzik\u00e1lnymi filtrami (vibra\u010dn\u00e9 odizolovanie), redundantn\u00fdmi modalitami (radar+kamera) a logikou eskal\u00e1cie (ak je neistota vysok\u00e1, zn\u00ed\u017e r\u00fdchlos\u0165, zv\u00fd\u0161 separ\u00e1ciu).<\/p>\n<h2>Praktick\u00e1 v\u00fdvojov\u00e1 a valida\u010dn\u00e1 pipeline<\/h2>\n<ol>\n<li><strong>Po\u017eiadavky a hazardy<\/strong>: definuj MMD\/TCPA limity, latencie, triedy cie\u013eov; vytvor hazard log a bezpe\u010dnostn\u00e9 ciele.<\/li>\n<li><strong>D\u00e1ta<\/strong>: zozbieraj re\u00e1lne sekvencie a syntetiku; priprav <em>scenario library<\/em> so \u0161tandardizovan\u00fdmi skriptmi.<\/li>\n<li><strong>Modely<\/strong>: navrhni supervised detekciu a self-supervised h\u013abku; pridaj anom\u00e1lne detektory a kalibr\u00e1ciu neistoty.<\/li>\n<li><strong>Tr\u00e9ning<\/strong>: strat\u00e9giu triednej nerovnov\u00e1hy, curriculum pod\u013ea vzdialenosti objektu, <em>mixed precision<\/em>, valid\u00e1cia na OOD setoch.<\/li>\n<li><strong>Integr\u00e1cia<\/strong>: ROS2 graf, \u010dasov\u00e9 zna\u010dky, synchroniz\u00e1cia senzorov, hard-real-time pre pl\u00e1novanie.<\/li>\n<li><strong>HIL\/SIL<\/strong>: v slu\u010dke s autopilotom, injekcie zlyhan\u00ed senzorov a komunika\u010dn\u00fdch v\u00fdpadkov.<\/li>\n<li><strong>Letov\u00e9 sk\u00fa\u0161ky<\/strong>: stup\u0148ovit\u00e9 nasadzovanie (static track\u2192captive carry\u2192pomalej\u0161\u00ed let\u2192re\u00e1lne scen\u00e1re), telemetria a <em>post-mortem<\/em> anal\u00fdzy.<\/li>\n<\/ol>\n<h2>Energetika a termika<\/h2>\n<p>V\u00fdpo\u010dtov\u00e1 n\u00e1ro\u010dnos\u0165 DAA v\u00fdznamne vpl\u00fdva na v\u00fddr\u017e letu. Optimalizuj pipeline (zn\u00ed\u017eenie rozl\u00ed\u0161enia pri n\u00edzkej r\u00fdchlosti, adapt\u00edvne FPS pod\u013ea rizika, preru\u0161ovan\u00e9 radarov\u00e9 d\u00e1vky), pou\u017e\u00edvaj jadra NPU\/DSP na n\u00edzkej \u00farovni a zabezpe\u010d termick\u00e9 riadenie (heat-spreaders, pr\u00fadenie vzduchu), aby sa predi\u0161lo throttlingu.<\/p>\n<h2>Etick\u00e9 a zodpovedn\u00e9 AI aspekty<\/h2>\n<p>Transparentnos\u0165 rozhodovania (saliency mapy, vysvetlite\u013en\u00e9 pr\u00edznaky) zvy\u0161uje d\u00f4veru. Minimalizuj <em>dataset bias<\/em> (r\u00f4zne regi\u00f3ny, ro\u010dn\u00e9 obdobia, typy objektov) a zav\u00e1dzaj <em>human-in-the-loop<\/em> pre kritick\u00e9 kroky valid\u00e1cie. Loguj d\u00f4vody man\u00e9vrov a dovo\u013e post-hoc audit.<\/p>\n<h2>Bud\u00face smerovania<\/h2>\n<p>O\u010dak\u00e1va sa v\u00e4\u010d\u0161ie vyu\u017eitie vizu\u00e1lno-inertnej SLAM h\u013abky so self-supervised doh\u013eadom, multimod\u00e1lnych transformerov (kamera-radar-IMU) s kalibrovanou neistotou a form\u00e1lnych <em>learning-enabled<\/em> garanci\u00ed (kombin\u00e1cia u\u010denia a bari\u00e9rov\u00fdch funkci\u00ed). V rojov\u00fdch aplik\u00e1ci\u00e1ch porastie v\u00fdznam distribuovanej predikcie a komunika\u010dne-efekt\u00edvnych reprezent\u00e1ci\u00ed hrozieb.<\/p>\n<p>DAA zalo\u017een\u00e9 na u\u010den\u00ed s u\u010dite\u013eom a bez u\u010dite\u013ea umo\u017e\u0148uje v\u00fdrazne prekro\u010di\u0165 limity tradi\u010dn\u00fdch heurist\u00edk. Supervised met\u00f3dy pon\u00fakaj\u00fa presn\u00fa percepciu zn\u00e1mych hrozieb, unsupervised vrstva str\u00e1\u017ei nezn\u00e1me a men\u00ed syst\u00e9m na <em>self-aware<\/em> modul schopn\u00fd signalizova\u0165 neistotu a anom\u00e1lie. V spojen\u00ed s validovan\u00fdm pl\u00e1novan\u00edm, runtime assurance a disciplinovanou verifik\u00e1ciou je mo\u017en\u00e9 dosiahnu\u0165 bezpe\u010dn\u00fa auton\u00f3miu v komplexn\u00fdch a dynamick\u00fdch podmienkach skuto\u010dn\u00e9ho sveta.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Detekcia a vyh\u00fdbanie cez ML: tr\u00e9ning, valid\u00e1cia a predikcia dr\u00e1h dynamick\u00fdch objektov pre bezpe\u010dn\u00e9 man\u00e9vrovanie.<\/p>\n","protected":false},"author":44,"featured_media":88685,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2629],"tags":[2479,2480,2481,2482,2258,2483,2484,2119],"class_list":["post-48685","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-drony","tag-daa-ucenim","tag-dataset","tag-detekcia","tag-generalizacia","tag-kolizie","tag-predikcia-trajektorii","tag-supervised-unsupervised","tag-validacia"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v27.5 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Detekcia a vyh\u00fdbanie (DAA): Vyu\u017eitie hlbok\u00e9ho u\u010denia pre prevenciu kol\u00edzi\u00ed - Podnik\u00e1m, lietam a relaxujem<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/\" \/>\n<meta property=\"og:locale\" content=\"sk_SK\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Detekcia a vyh\u00fdbanie (DAA): Vyu\u017eitie hlbok\u00e9ho u\u010denia pre prevenciu kol\u00edzi\u00ed - Podnik\u00e1m, lietam a relaxujem\" \/>\n<meta property=\"og:description\" content=\"Detekcia a vyh\u00fdbanie cez ML: tr\u00e9ning, valid\u00e1cia a predikcia dr\u00e1h dynamick\u00fdch objektov pre bezpe\u010dn\u00e9 man\u00e9vrovanie.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/\" \/>\n<meta property=\"og:site_name\" content=\"Podnik\u00e1m, lietam a relaxujem\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/vrtulniky\/\" \/>\n<meta property=\"article:published_time\" content=\"2025-08-18T15:52:04+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.vrtulniky.sk\/news\/wp-content\/uploads\/2022\/08\/vrtulniky-sk.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"1400\" \/>\n\t<meta property=\"og:image:height\" content=\"900\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"author\" content=\"J\u00e1n Ga\u0161par\u00edk\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Autor\" \/>\n\t<meta name=\"twitter:data1\" content=\"J\u00e1n Ga\u0161par\u00edk\" \/>\n\t<meta name=\"twitter:label2\" content=\"Predpokladan\u00fd \u010das \u010d\u00edtania\" \/>\n\t<meta name=\"twitter:data2\" content=\"11 min\u00fat\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\\\/\"},\"author\":{\"name\":\"J\u00e1n Ga\u0161par\u00edk\",\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/#\\\/schema\\\/person\\\/d68e7bbcdd214c81eeacaeea7b95cfd1\"},\"headline\":\"Detekcia a vyh\u00fdbanie (DAA): Vyu\u017eitie hlbok\u00e9ho u\u010denia pre prevenciu kol\u00edzi\u00ed\",\"datePublished\":\"2025-08-18T15:52:04+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\\\/\"},\"wordCount\":2143,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/wp-content\\\/uploads\\\/2025\\\/12\\\/vzdelavanie-vysoka-skola-8685.jpg\",\"keywords\":[\"DAA u\u010den\u00edm\",\"dataset\",\"detekcia\",\"generaliz\u00e1cia\",\"kol\u00edzie\",\"predikcia trajekt\u00f3ri\u00ed\",\"supervised\\\/unsupervised\",\"valid\u00e1cia\"],\"articleSection\":[\"Drony\"],\"inLanguage\":\"sk-SK\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\\\/#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\\\/\",\"url\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\\\/\",\"name\":\"Detekcia a vyh\u00fdbanie (DAA): Vyu\u017eitie hlbok\u00e9ho u\u010denia pre prevenciu kol\u00edzi\u00ed - Podnik\u00e1m, lietam a relaxujem\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/wp-content\\\/uploads\\\/2025\\\/12\\\/vzdelavanie-vysoka-skola-8685.jpg\",\"datePublished\":\"2025-08-18T15:52:04+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\\\/#breadcrumb\"},\"inLanguage\":\"sk-SK\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"sk-SK\",\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\\\/#primaryimage\",\"url\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/wp-content\\\/uploads\\\/2025\\\/12\\\/vzdelavanie-vysoka-skola-8685.jpg\",\"contentUrl\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/wp-content\\\/uploads\\\/2025\\\/12\\\/vzdelavanie-vysoka-skola-8685.jpg\",\"width\":1248,\"height\":832,\"caption\":\"Detekcia a vyh\u00fdbanie (DAA): Vyu\u017eitie hlbok\u00e9ho u\u010denia pre prevenciu kol\u00edzi\u00ed\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Detekcia a vyh\u00fdbanie (DAA): Vyu\u017eitie hlbok\u00e9ho u\u010denia pre prevenciu kol\u00edzi\u00ed\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/#website\",\"url\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/\",\"name\":\"Podnik\u00e1m, lietam a relaxujem\",\"description\":\"Vrtu\u013en\u00edky.sk\",\"publisher\":{\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"sk-SK\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/#organization\",\"name\":\"Podnik\u00e1m, lietam a relaxujem\",\"url\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"sk-SK\",\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/wp-content\\\/uploads\\\/2022\\\/08\\\/news-vrtulniky-sk-logo-e1660318023553.png\",\"contentUrl\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/wp-content\\\/uploads\\\/2022\\\/08\\\/news-vrtulniky-sk-logo-e1660318023553.png\",\"width\":201,\"height\":200,\"caption\":\"Podnik\u00e1m, lietam a relaxujem\"},\"image\":{\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/www.facebook.com\\\/vrtulniky\\\/\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/#\\\/schema\\\/person\\\/d68e7bbcdd214c81eeacaeea7b95cfd1\",\"name\":\"J\u00e1n Ga\u0161par\u00edk\",\"url\":\"https:\\\/\\\/www.vrtulniky.sk\\\/news\\\/author\\\/jan-gasparik\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Detekcia a vyh\u00fdbanie (DAA): Vyu\u017eitie hlbok\u00e9ho u\u010denia pre prevenciu kol\u00edzi\u00ed - Podnik\u00e1m, lietam a relaxujem","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/","og_locale":"sk_SK","og_type":"article","og_title":"Detekcia a vyh\u00fdbanie (DAA): Vyu\u017eitie hlbok\u00e9ho u\u010denia pre prevenciu kol\u00edzi\u00ed - Podnik\u00e1m, lietam a relaxujem","og_description":"Detekcia a vyh\u00fdbanie cez ML: tr\u00e9ning, valid\u00e1cia a predikcia dr\u00e1h dynamick\u00fdch objektov pre bezpe\u010dn\u00e9 man\u00e9vrovanie.","og_url":"https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/","og_site_name":"Podnik\u00e1m, lietam a relaxujem","article_publisher":"https:\/\/www.facebook.com\/vrtulniky\/","article_published_time":"2025-08-18T15:52:04+00:00","og_image":[{"width":1400,"height":900,"url":"https:\/\/www.vrtulniky.sk\/news\/wp-content\/uploads\/2022\/08\/vrtulniky-sk.jpg","type":"image\/jpeg"}],"author":"J\u00e1n Ga\u0161par\u00edk","twitter_card":"summary_large_image","twitter_misc":{"Autor":"J\u00e1n Ga\u0161par\u00edk","Predpokladan\u00fd \u010das \u010d\u00edtania":"11 min\u00fat"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/#article","isPartOf":{"@id":"https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/"},"author":{"name":"J\u00e1n Ga\u0161par\u00edk","@id":"https:\/\/www.vrtulniky.sk\/news\/#\/schema\/person\/d68e7bbcdd214c81eeacaeea7b95cfd1"},"headline":"Detekcia a vyh\u00fdbanie (DAA): Vyu\u017eitie hlbok\u00e9ho u\u010denia pre prevenciu kol\u00edzi\u00ed","datePublished":"2025-08-18T15:52:04+00:00","mainEntityOfPage":{"@id":"https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/"},"wordCount":2143,"commentCount":0,"publisher":{"@id":"https:\/\/www.vrtulniky.sk\/news\/#organization"},"image":{"@id":"https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/#primaryimage"},"thumbnailUrl":"https:\/\/www.vrtulniky.sk\/news\/wp-content\/uploads\/2025\/12\/vzdelavanie-vysoka-skola-8685.jpg","keywords":["DAA u\u010den\u00edm","dataset","detekcia","generaliz\u00e1cia","kol\u00edzie","predikcia trajekt\u00f3ri\u00ed","supervised\/unsupervised","valid\u00e1cia"],"articleSection":["Drony"],"inLanguage":"sk-SK","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/#respond"]}]},{"@type":"WebPage","@id":"https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/","url":"https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/","name":"Detekcia a vyh\u00fdbanie (DAA): Vyu\u017eitie hlbok\u00e9ho u\u010denia pre prevenciu kol\u00edzi\u00ed - Podnik\u00e1m, lietam a relaxujem","isPartOf":{"@id":"https:\/\/www.vrtulniky.sk\/news\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/#primaryimage"},"image":{"@id":"https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/#primaryimage"},"thumbnailUrl":"https:\/\/www.vrtulniky.sk\/news\/wp-content\/uploads\/2025\/12\/vzdelavanie-vysoka-skola-8685.jpg","datePublished":"2025-08-18T15:52:04+00:00","breadcrumb":{"@id":"https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/#breadcrumb"},"inLanguage":"sk-SK","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/"]}]},{"@type":"ImageObject","inLanguage":"sk-SK","@id":"https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/#primaryimage","url":"https:\/\/www.vrtulniky.sk\/news\/wp-content\/uploads\/2025\/12\/vzdelavanie-vysoka-skola-8685.jpg","contentUrl":"https:\/\/www.vrtulniky.sk\/news\/wp-content\/uploads\/2025\/12\/vzdelavanie-vysoka-skola-8685.jpg","width":1248,"height":832,"caption":"Detekcia a vyh\u00fdbanie (DAA): Vyu\u017eitie hlbok\u00e9ho u\u010denia pre prevenciu kol\u00edzi\u00ed"},{"@type":"BreadcrumbList","@id":"https:\/\/www.vrtulniky.sk\/news\/detekcia-a-vyhybanie-daa-vyuzitie-hlbokeho-ucenia-pre-prevenciu-kolizii\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.vrtulniky.sk\/news\/"},{"@type":"ListItem","position":2,"name":"Detekcia a vyh\u00fdbanie (DAA): Vyu\u017eitie hlbok\u00e9ho u\u010denia pre prevenciu kol\u00edzi\u00ed"}]},{"@type":"WebSite","@id":"https:\/\/www.vrtulniky.sk\/news\/#website","url":"https:\/\/www.vrtulniky.sk\/news\/","name":"Podnik\u00e1m, lietam a relaxujem","description":"Vrtu\u013en\u00edky.sk","publisher":{"@id":"https:\/\/www.vrtulniky.sk\/news\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.vrtulniky.sk\/news\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"sk-SK"},{"@type":"Organization","@id":"https:\/\/www.vrtulniky.sk\/news\/#organization","name":"Podnik\u00e1m, lietam a relaxujem","url":"https:\/\/www.vrtulniky.sk\/news\/","logo":{"@type":"ImageObject","inLanguage":"sk-SK","@id":"https:\/\/www.vrtulniky.sk\/news\/#\/schema\/logo\/image\/","url":"https:\/\/www.vrtulniky.sk\/news\/wp-content\/uploads\/2022\/08\/news-vrtulniky-sk-logo-e1660318023553.png","contentUrl":"https:\/\/www.vrtulniky.sk\/news\/wp-content\/uploads\/2022\/08\/news-vrtulniky-sk-logo-e1660318023553.png","width":201,"height":200,"caption":"Podnik\u00e1m, lietam a relaxujem"},"image":{"@id":"https:\/\/www.vrtulniky.sk\/news\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/vrtulniky\/"]},{"@type":"Person","@id":"https:\/\/www.vrtulniky.sk\/news\/#\/schema\/person\/d68e7bbcdd214c81eeacaeea7b95cfd1","name":"J\u00e1n Ga\u0161par\u00edk","url":"https:\/\/www.vrtulniky.sk\/news\/author\/jan-gasparik\/"}]}},"_links":{"self":[{"href":"https:\/\/www.vrtulniky.sk\/news\/wp-json\/wp\/v2\/posts\/48685","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.vrtulniky.sk\/news\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.vrtulniky.sk\/news\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.vrtulniky.sk\/news\/wp-json\/wp\/v2\/users\/44"}],"replies":[{"embeddable":true,"href":"https:\/\/www.vrtulniky.sk\/news\/wp-json\/wp\/v2\/comments?post=48685"}],"version-history":[{"count":0,"href":"https:\/\/www.vrtulniky.sk\/news\/wp-json\/wp\/v2\/posts\/48685\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.vrtulniky.sk\/news\/wp-json\/wp\/v2\/media\/88685"}],"wp:attachment":[{"href":"https:\/\/www.vrtulniky.sk\/news\/wp-json\/wp\/v2\/media?parent=48685"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.vrtulniky.sk\/news\/wp-json\/wp\/v2\/categories?post=48685"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.vrtulniky.sk\/news\/wp-json\/wp\/v2\/tags?post=48685"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}