Namhlanje thina Thetha ngamaqela adluleyo . Kuya kubakho ababini kubo kule habrapost. Eyokuqala yiLabra yokuqalisa, ephuhlisa isisombululo sokubeka iliso kwimveliso yabasebenzi. Isibini - ngenkqubo yokuqaphela ubuso kwii-turnstiles.
Ifoto: /unsplash.com
ILabra iya kwandisa njani imveliso
Ukukhula kwemveliso kwiimarike zaseNtshona kuye kwacotha. Ngu McKinsey, ekuqaleni kweminyaka yoo-2,4 eli nani laliyi-2010%. Kodwa phakathi kuka-2014 kunye no-0,5 yehla ukuya kwi-2%. Abahlalutyi baqaphela ukuba imeko ayikatshintshi ukususela ngoko. Kodwa kukho uluvo lokuba iinkqubo zobuntlola ezenziweyo ziya kunceda ukusombulula ingxaki. Ngoncedo lweenkqubo ze-AI, ukukhula kwemveliso kulindeleke ukuba kubuyele kwi-XNUMX% kwiminyaka elishumi. Ii-algorithms ezihlakaniphile ziya kunceda ukwenza imisebenzi yesiqhelo kunye nokwandisa iinkqubo zokusebenza.
Uphando kwezi ndawo sele luqhutywa ziingcali ezivela , iinjineli kunye nabameli . Umbono womatshini uya kudlala indima ebalulekileyo ekwandiseni ukukhula kwemveliso. Itekhnoloji isetyenziselwa ukuvavanya ngokuzimeleyo indawo yokusebenza kunye nokusebenza komqeshwa. Izisombululo ezinjalo sele ziphunyezwa ziinkampani zaseNtshona - umzekelo, ΠΈ .
Iinkampani zaseRashiya zikwaphuhlisa izisombululo zokuvavanya imveliso yabasebenzi. Ngokomzekelo, i-Labra yokuqalisa, eyadlula yethu . Iinjineli zenza inkqubo yokujonga ividiyo ngenethiwekhi ye-neural eqaphela izenzo zabasebenzi beshishini kwaye iyenza icace gca indlela abalichitha ngayo ixesha labo lokusebenza.
Indlela esebenza ngayo inkqubo. I-Labra inokusebenza kulo naliphi na ishishini elinomatshini okanye ngoomatshini-osebenza ngezandla ababasebenzi abangaphezulu kwe-15. Ngoncedo lweekhamera, wenza into ebizwa ngokuba - oko kukuthi, irekhoda yonke into eyenzekayo ngexesha lokutshintsha. Ngokubanzi, i-algorithm ibonakala ngolu hlobo:
- Inkqubo ibamba umfanekiso kwaye iphawule imisebenzi yomsebenzi;
- I-algorithm yokufunda ngomatshini ihlalutya ividiyo;
- I-algorithm ke ivelisa ifoto yosuku lokusebenza;
- Okulandelayo, i-analytics ibalwa ngokuzenzekelayo;
- I-Labra ivelisa ingxelo yokugqibela kunye neengcebiso eziza kwandisa ukhuseleko kwishishini kunye nokwandisa izixhobo zayo.
Ngubani okwiqela? Isiqalo sinabasebenzi abasibhozo: umphathi kunye nomseki, abaphuhlisi ababini, iingcali ezintathu zemigangatho yabasebenzi. Kukwakho nomphathi wenkonzo yabathengi kunye nomgcini zincwadi zemali. Abanye babo badibanisa umsebenzi weprojekthi kunye nezifundo zaseyunivesithi. Ke ngoko, wonke umntu ujonga ukugqitywa kwemisebenzi kunye nemihla ebekiweyo ngokuzimeleyo. Nangona kunjalo, iqela libamba iindibano kabini ngeveki ukuxoxa ngenkqubela phambili kunye nezicwangciso zophuhliso.
Amathemba. Ekuqaleni kukaSeptemba, ukuqaliswa kwabonisa iprojekthi yayo . Apho, iinjineli zabonisa amandla emveliso. I-Labra iceba ukuqhubela phambili ukukhuthaza isisombululo kwaye isebenza kwithemba lentsebenziswano kunye namashishini kweli lizwe.
I-O.VISION iya kukunceda ulahle izitshixo kunye nokupasa
Kwi-2017, uPhononongo lweTekhnoloji yeMIT ukuqondwa kobuso kwi-10 ephezulu yetekhnoloji yempumelelo. Esi sigqibo ngokuyinxenye kungenxa yokusetyenziswa ngokubanzi kweenkqubo ezinjalo. Ngokukodwa, banokuthatha indawo yezitshixo eziqhelekileyo kwaye badlule xa bengena kwisakhiwo - umzekelo, inani leebhanki zaseRashiya sele liphunyezwe uphuhliso olufanayo. Abadlali abatsha nabo babonakala kwiimarike, umzekelo, isiqalo siphuhlisa isisombululo esifanayo . Iqela lenza inkqubo yokufikelela ngaphandle koqhagamshelwano kwi-turnstiles enokufakwa kwimizuzu engama-30.
Indlela esebenza ngayo inkqubo. Uphuhliso yisoftware kunye nehardware complex efakwe kwindawo yokukhangela. Isekwe kwiinethiwekhi ezintlanu ze-neural ezenza isakhelo ngasinye ukusuka kwikhamera yenkqubo yebhayometriki. Ababhali bathi ukusetyenzwa komfanekiso omnye kuthatha ngaphantsi kwe-200 millisecond (malunga nezakhelo ezintlanu ngesekhondi). Iqela libhala zonke ii-algorithms zokuqaphela kunye ne-interfaces ngokuzimeleyo-abaphuhlisi abasebenzisi izisombululo zobunikazi. Qeqesha uthungelwano lwe-neural usebenzisa .
Ukusetyenzwa kwedatha kwenzeka ekuhlaleni. Le ndlela yokwandisa ukhuseleko lwedatha yebhayometriki yobuqu. I-hardware iquka ibhodi ye-Jetson TX1 evela kwi-Nvidia, eyenzelwe izixhobo ezizimeleyo. Inkqubo yebhayometriki ikwaqulethe isekethe edibeneyo yoyilo lwayo lokulawula i-turnstiles kunye nokudibanisa kunye .
Ifoto: /unsplash.com
Abasebenzi bokuqala. Intloko yenkampani ithi ukhetho lwenziwa ngokomgaqo: abaviwa abangama-60 kwindawo enye. Le fomati yasivumela ukuba sifumane abona bantu banetalente. Okwangoku, abadwelisi benkqubo abaliqela basebenza kule projekthi, inoxanduva lokufunda ngoomatshini kunye nekhowudi yeenkqubo ezifakiweyo. Kukho kwakhona umphuhlisi we-backend, ingcali yokhuseleko lolwazi kunye nomyili. Abanye babasebenzi ngabafundi abadibanisa umsebenzi kunye nesidanga se-masters.
Amathemba. Izisombululo zanamhlanje ifakwe kumzi-mveliso wekofu omkhulu eYurophu. Imveliso iphinda ilungiselelwe ukuqaliswa kwelinye lamaziko empilo eSt. Petersburg kunye neYunivesithi yasePolytechnic. Mhlawumbi kwixesha elizayo i-O.VISION iya kufakwa kwiYunivesithi ye-ITMO. Intloko yenkampani ithi sele sele ixoxisana neenkampani zaseRashiya: iGazprom Neft, iBeeline, iRostelecom kunye neRussian Railways. Kwixesha elizayo, siya kungena kwiimarike zangaphandle.
Malunga nezinye iiprojekthi ze-accelerator:
Izinto eziphathekayo malunga nomsebenzi weYunivesithi ye-ITMO:
umthombo: www.habr.com