Uhlelo lokuqaphela ubuso ngosizo lwezinhlelo zokubhekwa kwevidiyo. I-Algorithm yokusesha ubuso

Isofthiwe ne-hardware yezinhlelo zokuphepha ezididiyelwe zanamuhla ziyakwazi ukuxazulula izinkinga zanoma iyiphi inkimbinkimbi kuzo zonke izinhlobo zezinto zezimboni, ezenhlalo nezokufuywayo. Amathuluzi abaluleke kakhulu wezinhlelo zokuphepha yizinhlelo zokubhekwa kwevidiyo, futhi izidingo zokusebenza kwesigaba ziyaqhubeka zikhula.

Izinhlelo zokuphepha ezihlanganisiwe

Isiteji esisodwa sezinhlelo zokuvikeleka ezididiyelwe zihlanganisa amamojula wokuphepha nemishini yomlilo, ukulawula nokufinyelela kokufinyelela, ividiyo yokubhekwa noma ithelevishini yezokuphepha (i-COT). Imisebenzi yalezi zinsuku kuze kube yamuva zazinqunywa ekuqaphelweni kwevidiyo nokuqoshwa kwesimo esiteshini nasendaweni eseduze, ukugcinwa kwemininingwane kanye nokugcinwa kwedatha. Izinhlelo zevidiyo zasendulo zinezinkinga eziningi eziphawulekayo:

• Isici somuntu. Ukusebenza okungalungile kopharetha uma kusakaza ulwazi oluningi.
• Ukungenakwenzeka kokungenelela kokuhlinzwa, ukuhlaziywa okungakazelelwe.
• Izindleko eziqakathekile zesikhathi sokuthola nokukhomba umcimbi.

Ukuthuthukiswa kobuchwepheshe bedijithali kwaholela ekwakhiweni kwezinhlelo ze "smart" ezizenzakalelayo.

Amandla Ngobuhlakani

Isimiso esiyisisekelo se-smart intelligent system surveillance system yi-video analytics - ubuchwepheshe obusekelwe ezindleleni nasezinqumweni zokubona isithombe nokucubungula isithombe, ukuqoqwa kwedatha okuzenzakalelayo ngenxa yokuhlaziywa kwevidiyo. Imishini enjalo ngaphandle kokungenelela komuntu iyakwazi ukubona nokulandelela ngesikhathi sangempela izinhloso ezibekwe (imoto, iqembu labantu), izimo ezingase zibe yingozi (intuthu, umlilo, ukuphazanyiswa okungagunyaziwe nokusebenza kwamakhamera wevidiyo), imicimbi ehleliwe kanye nokukhishwa okwesikhashana i-alamu. Ngenxa yokuhlunga kwedatha yevidiyo engekho nentshisekelo, umthwalo kumashaneli wokuxhumana kanye nedatha yokugcina ingosi yomlando iyancipha kakhulu.

Izindlela ezidume kakhulu ze-analytics yevidiyo yindlela yokubona ubuso. Kuncike emisebenzini eyenziwe kanye nemisebenzi eyabelwe, izidingo ezithile zifakwe kule mishini.

I-Firmware

Ukuze kusebenze uhlelo olusebenzayo, izinhlobo eziningana zekhamera ze-IP ezinezici ezihlukahlukene zokusebenza zisetshenziswa. Ukutholwa kwento ensimini elawulwayo kulungiswe ngamakhamera we-panoramic ngokuxazulula kwe-1 Mp kanye nobude obugxile bu-1 mm futhi buqondiswa ngamadivayisi wokuskena. Lezi amakhamera aphezulu kakhulu (kusuka ku-2Mp, kusuka ku-2mm), okukhiqiza ukuqashelwa ngamasu alula (3-4 imingcele). Ukubona into isebenzisa amakhamera enemfanelo emihle yesithombe, eyanele ukusetshenziswa kwezinhlelo eziyinkimbinkimbi (kusuka ku-5 Mp, 8-12 mm). Isofthiwe ethandwa kakhulu ekuqaphelisweni ubuso "Ubuso Bokuqonda" (umthuthukisi - inkampani yokulawula indlu), umqondisi we-Face (inkampani "Synesis" ne-VOCORD FaceControl (VOCORD) ibonisa:

• Amathuba aphezulu okukhomba into (kufika ku 99%).
• Ukusekela ama-angles amaningi wokujikeleza kwekhamera.
• Ukuthi kungenzeka ukuhlukanisa abantu ngisho nasesisindweni esinabantu abahamba ngezinyawo.
• Ukuhluka kokulungiselela imibiko yokuhlaziya.

Izinsisekelo zokuqaphela iphethini

Noma yiziphi izinhlelo zokuqashelwa kwe-biometric zisekelwe ekuboniseni izincwadi eziphathelene nomzimba womuntu ofundwa kwiphrofayli ethile echazwe ngaphambilini. Ukuskena kwenzeka ngesikhathi sangempela. Ikhamera ye-IP isakaza ukusakazwa kwevidiyo ku-terminal, futhi uhlelo lokuqaphela ubuso lubona ukuxhumana kwesithombe ezithombeni ezigcinwe ku-database. Kunezindlela ezimbili eziyinhloko. Iyokuqala isekelwe ezimisweni eziqinile: imiphumela yokucubungula imingcele ye-biometric yenza i-sampula ye-elektroniki ngesimo senombolo ehlukile ehambisana nomuntu othile. Indlela yesibili imodeli "indlela" yabantu futhi ibonakala ngokuzifundela nokuqina. Ukukhomba komuntu ngokwemifanekiso yevidiyo kwenzeka ngokucubungula ukuguqulwa kweminyaka kanye nezinye izici (ukutholakala kwekhanda, isilevu noma amadevu, izibuko). Lolu buchwepheshe bukuvumela ukuba usebenze ngisho nezithombe ezindala futhi, uma kunesidingo, nge-X-ray.

I-Algorithm yokusesha ubuso

Indlela ejwayelekile kakhulu yokubona ubuso isebenzisa ukuhamba kweHaar (isethi yamaski). I-mask iyiwindwangu engxenyeni enenhlanganisela ehlukene yamacici amhlophe namnyama.

Indlela yokwenza lolu hlelo kanje: ifreyimu yevidiyo ihlanganiswe nesethi yamaski, kanye nemiphumela ye-convolution (ukubalwa kwamaphikseli abanjwe emikhakha emhlophe namnyama) umehluko ubalwa, uma kuqhathaniswa nenani elithile lomngcele.

Ukuze uthuthukise umsebenzi we-classifier, omuhle (ozimele, lapho ubuso bomuntu bukhona khona) futhi okungalungile (ngaphandle kwalezo) amasampula wokuqeqeshwa adalwa. Esikhathini sokuqala, umphumela we-convolution ungaphezu kwe-threshold value, esimweni sesibili sincane. Umtshina wabantu abanephutha elivumelekile unquma inani le-convolutions yazo zonke izikhukhula futhi, uma isibalo sinqunywe, sibonisa ukuthi kukhona khona abantu abakulesi sakhiwo.

Ubuchwepheshe bokuqaphela

Ngemuva kokuthola nokwasekhaya, ukukhanya nokulinganisa kwe-geometri yesithombe kwenzeka esiteji sokuqala. Izenzo ezengeziwe - ukubalwa kwezici nokuhlonza - kungenziwa ngezindlela ezihlukahlukene.

Uma uskena ubuso obugcwele ekamelweni elinokukhanyisa okuhle kakhulu, imiphumela emihle iboniswa yi-algorithms esebenza ngezithombe ze-2D. Ukuhlaziya amaphuzu ayingqayizivele nama-distances phakathi kwabo, uhlelo lokuqaphela ubuso lubonisa iqiniso lokuhlonza ngama-coefficients of the umehluko phakathi kwesithombe sokuthi "bukhoma" nesifanekiso esibhalisiwe.

Ubuchwepheshe obuthathu bubukeka buphikisana nezinguquko ekukhanyeni okukhanyayo, ukuphambuka okuvumelekile kusuka engxenyeni yangaphambili kuya kuma-degree angu-45. Lapha, akuyona amaphuzu kanye nemigqa kuphela ehlaziywayo, kepha futhi izindawo zokuhlala (ukuvuthwa, iphrofayli), i-metric yebanga eliphakathi kwabo. Ukuze kusetshenziswe izilungiswa ezinjalo, izinga eliphezulu lokurekhoda kwevidiyo nomvuthwandaba ongama-200 ozimele ngomzuzwana kuyadingeka. Isisekelo sesistimu siqukethe amakhamera we-stereo ane-matrix yama-megapixel angu-5, isinqumo esiphakeme se-optical kanye nephutha lokuvumelanisa lokunciphisa. Ngaphezu kwalokho, zixhunyiwe ngekhebula elikhethiwe lokwenza isikhathi sokudlulisela ukuxilonga kokuvumelanisa.

Umbuso wezimakethe zesimanje

Izindlela zokuqala zokulawulwa kwe-biometric, ngenxa yezindleko zabo eziphakeme, zenzelwe kuphela izakhiwo zempi zombuso futhi kuphela phakathi nawo-1990 zathola izinhlangano zezohwebo. Ukuthuthukiswa okusheshayo kwezobuchwepheshe kanye ne-microprocessor ubuchwepheshe kuye kwavumela ukwandisa ukunemba kwezinhlelo nokwandisa ububanzi besicelo sabo. Emakethe yezwe lethu eliholayo izikhundla kubakhiqizi bezokuphepha baseMelika naseWest Europe. Umholi wezokuthengisa yizinto zokusebenza zezinkampani ZN Vision Technologies kanye neMiboniso. Othembisayo kakhulu phakathi kwabathuthukisi baseRussia ucwaningo kanye nemikhiqizo yeVocord, NTechLab, Soling, VizhnLabs kanye namaqembu e-MDG, okunye, phakathi kwezinye izinto, okubandakanyeka ekuguquleni izimo zakwamanye amazwe ezindaweni zesiRashiya.

I-Computer Face Control

Insimu ebanzi kakhulu yokusetshenziswa kokungabonakali okungaxhunyiwe kuxhumana nokulwa nobuphekula nobugebengu. Isithombe sesimo sobugebengu sigcinwa ku-database. Ezindaweni zokuxubana kwabantu abaningi (izindiza zezindiza, iziteshi zesitimela, ezitolo zezitolo, izakhiwo zezemidlalo), ukusakazwa kwangempela kwabantu kuboniswa ukuhlonza abantu abafunayo.

I-sphere esilandelayo yizinhlelo zokulawula ukufinyelela: isampula yesithombe esithombeni ekudluleleni kwe-elektroniki kuqhathaniswa nemodeli etholakala ngenxa yokucubungula idatha kusuka kumakhamera wevidiyo. Inqubo yenzeke ngokushesha, ngaphandle kokwenza noma yiziphi izenzo ezengeziwe (ngokuphambene nokuskena i-retina yehlo noma i-dactyloscopy).

Enye imboni ekhula ngokushesha iyamaketha. I- billboard esebenzisanayo, ngokuskena ubuso bomuntu, inquma ubulili nobudala bakhe, ibheka kuphela lezo zikhangiso ezingase zithande iklayenti.

Amathrendi namathemba okuthuthukiswa

Izindlela zokuqashelwa kwabantu emkhakheni webhange ziyadingeka kakhulu. Ngenxa yonyaka odlule, emva kokufakwa kwamakhamera wevidiyo ama-50,000 ahlakaniphile emahhovisi awo, i-Post Bank Bank yakwazi ukugcina izigidi zama-ruble ngokuvimbela ukukhwabanisa kumagcekeni esikweletu kanye nokukhokha. Ochwepheshe bathi i-2021 inethiwekhi engqalasizinda edingekayo izokwakhiwa futhi noma yikuphi ukusebenza kuma-ATM kuzokwenzeka kuphela ngemuva kokubonakala kwesimo se-biometric ubuso bomthengi.

Kule minyaka eyishumi ezayo ubuchwepheshe obuphezulu buzovumela ukuvula inethiwekhi yezitolo ezigcwele ukuzisebenzela: umthengi udlula phambi kwezitolo, ukhetha izimpahla azithandayo futhi azishiye. Uhlelo lokubheka ubuso nemifanekiso luyonquma ukuthi ngubani umthengi, ukuthenga futhi uzokhipha imali yakhe i-akhawunti yakhe.

Umsebenzi uqhubeka ukudala izinhlelo zokuqaphela isimo sengqondo. Ukuhlaziywa kwemizwelo yabantu kuyoba kudingekile kuma-multimedia spheres: izithombe, i-cinematography, imboni yokudala imidlalo yekhompyutha.