Umbono ikhompyutha Modern. Imisebenzi kanye umbono computer nobuchwepheshe. UHlelo Computer Vision in Python

Kanjani ukufundisa ikhompyutha ukuthi iqonde ukuthi esithombeni noma izithombe? Lokhu kubonakala elula, kodwa i-computer lokhu nje matrix ehlanganisa amaqanda futhi ababevela ofuna kukhishwe ukwaziswa okubalulekile.

Uyini umbono khompyutha? Kuyinto ikhono "ukubona" kukhompyutha yakho

Vision - iwumthombo obalulekile ulwazi umuntu uyisebenzisa, sithola, ngokusho kombiko ezihlukahlukene, kusukela ngo-70 kuya ku-90% azo zonke imininingwane. Futhi-ke, uma sifuna ukwakha smart imoto, sidinga ukusebenzisa amakhono efanayo nekhompyutha.

Inkinga umbono computer kungenziwa okushiwo bha. Kuyini "ukubona"? Kuyaqondakala ukuthi lapho kukhona nje ngokubheka. Lokho waphetha umehluko umbono computer futhi umbono womuntu. Vision kithi - it is a umthombo wolwazi mayelana nezwe, kanye umthombo wokwaziswa imethrikhi - okungukuthi, ikhono ukuqonda amabanga nobukhulu obufanele.

Semantic ikhenela isithombe

Ukubheka isithombe, singakwazi ukuchaza ke ngenombolo izimfanelo, ngomqondo ongokomfanekiso, ukuba luthathe ukwaziswa semantic.

Ngokwesibonelo, ebuka lesi sithombe, singasho ukuthi ngaphandle. Kuyini sezimoto yedolobha. Lokho kukhona izimoto. singakwazi ukuqagela ukuthi lena South-East Asia ku ukucushwa kwe-wokwakha yokubhala. I isithombe Mao Zedong bavisise kwekutsi loko kwentiwe Beijing, futhi uma umuntu ebona ividiyo bukhoma noma wake waba khona, kwakuyobonisa ukuqagela ukuthi lokhu iyona edume Tiananmen Square.

What singasho kabanzi mayelana isithombe, ekubona? Singakwazi ukuthola ukuthi lingubani izinto esithombeni, ukusho, ukuthi kukhona abantu lapha eduze - kocingo. Lapha izambulela, amaphosta ukuthi sakhiwo. Lezi ziimbonelo zemihlobo amakilasi kuyinto izinto ezibaluleke kakhulu, okuyinto bekwi search for okwamanje.

Noma kunjalo singafunda ezinye izici noma izimfanelo lezinto. Ngokwesibonelo, lapha singakwazi ukuthola ukuthi lokhu akukona portrait we Chinese abavamile, okungukuthi, Mao Zedong.

Ngokusho imoto kungaba azimisele ukuthi into ihamba, futhi kunzima, esingavunyelwe zikhubazekile ngesikhathi ukunyakaza. Mayelana amafulegi kungashiwo ukuthi yini engathathwa ngokuthi, nazo ezihambayo, kodwa akuzona kanzima, njalo zikhubazekile. Futhi in the scene kukhona umoya, othatha izinhlayiyana kungenziwa kunqunywa ekuthuthukiseni amafulegi, futhi kungacacisa ukuthi ngisho kokuqondisa komoya, isibonelo, uphephula kusuka kwesokunxele kuya kwesokudla.

Yokuqhela ubude umbono ikhompyutha

Kubaluleke kakhulu iyona Imininingwane imethrikhi ngombono computer science. Lena zonke izinhlobo amabanga. Ngokwesibonelo, i-rover kubalulekile ngoba amaqembu bavela Umhlaba amaminithi angu-20 bese uphendula kakhulu. Ngakho, isixhumanisi lapho futhi emuva - imizuzu 40. Futhi uma senza uhlelo imiyalo ukunyakaza Zomhlaba, udinga ukuthatha lokhu ku-akhawunti.

ihlanganiswe ngempumelelo ubuchwepheshe umbono ikhompyutha isiqophi imidlalo. Ngokwe-video, ungakwazi ukwakha amamodeli ngakuthathu lezinto, abantu, nezithombe umsebenzisi angakwazi ukubuyisela onobuhle ngakuthathu amadolobha. Bese uhamba phezu kwabo.

umbono ikhompyutha - ibanga kunalokho ububanzi. It is eduze zithandelana neminye imikhakha yesayensi ahlukahlukene. Ingxenye umbono ikhompyutha It idlulisela processing image endaweni futhi ngezinye izikhathi inike umbono ikhompyutha, ngokomlando.

Ukuhlaziya, ukubonwa kwephethini - indlela eya ukudalwa ehlakaniphe kakhulu

Ake sihlole imiqondo le ngokwehlukile.

Image Icubungula - lena indawo algorithm, lapho input bese okukhipha - isithombe, futhi thina kufanele kuye ukwenza okuthile.

isithombe ukuhlaziywa - kuyinto endaweni umbono ikhompyutha, egxile ukusebenza isithombe ezimbili-ntathu futhi wenze tiphetfo.

Iphethini Recognition - isiyalo abstract zezibalo ukuthi ibona idatha ngesimo zithwala. Okungukuthi, emnyango - vector kanye sinokuthile ngayo. Lapho Vector liwukuthi siya- akubalulekile kangako ukwazi.

umbono Computer - ke ekuqaleni kwaba ukubuyisela nokwakheka iyimidwebo. Namuhla kule ndawo isibe ebanzi futhi lingahunyushwa ngalezi zindlela ukwamukela zonke eziphathekako ukwenza, esekelwe isithombe. Okungukuthi, kuba umsebenzi ingqondo yokufakelwa.

In parallel umbono ikhompyutha ensimini ehluke ngokuphelele, e geodesy, photogrammetry uye savela - yokukala ibanga phakathi izinto iyimidwebo.

Robots 'simbone'

Futhi ekugcineni - lena umshini umbono. Ngaphansi umbono umshini kusho umbono amarobhothi. Lokho kuyisinqumo ezinye izinkinga ukukhiqizwa. Singasho ukuthi umbono ikhompyutha - ingenye isayensi enkulu. Sihlanganisa ezinye isayensi ingxenye. Futhi lapho umbono ikhompyutha uthola yisiphi isicelo esithile, sikhule size sibe nombono umshini.

umbono Ikhompyutha esifundeni inobungako ngokoqobo. Lihlobene nokubhekana ubuso ezishintshayo ukukhiqizwa. Ngesikhathi amabhizinisi kangcono kakhulu ukuba ungene esikhundleni umsebenzi wezandla ngu umshini. Umshini engakutholi ukhathele, akozeli, ayenakho okungajwayelekile emsebenzini isimiso, yena-ke evuma ukuba asebenze izinsuku ezingu-365 ngonyaka. Ngakho, usebenzisa umsebenzi umshini, singathola yalokho yibhange ngesikhathi esithile, futhi kunzima kakhulu ezithakazelisayo. Yonke imisebenzi ezinokusetshenziswa ukusula umbono zekhompyutha. Futhi akukho lutho olungcono kuka ukuze ubone imiphumela ngokushesha isithombe kuphela ukubala esiteji.

Njengoba usuzokwenza emhlabeni ingqondo yokufakelwa

Plus endaweni - kunzima! Ingxenye enkulu yobuchopho obhekene umbono, futhi kunenkolelo yokuthi uma ukufundisa kukhompyutha yakho 'ukubona', okungukuthi, umbono ogcwele sokusebenzisa i-computer, kuba omunye izinjongo intelligence umnyuziki yokufakelwa. Uma singakwazi ukuxazulula inkinga ezingeni womuntu, okungenzeka kakhulu ngesikhathi esifanayo, sizobe ukuxazulula inkinga AI. Lokho kuhle kakhulu! Noma kahle, uma ubuka, "Terminator 2".

Kungani umbono - kunzima? Ngoba isithombe yezinto ezifanako ziyahlukahluka kuye kakhulu ezicini wangaphandle. Kuye into observation amaphuzu sehluke.

Ngokwesibonelo, omunye futhi isibalo esifanayo ethathwe-engeli lehlukene. Futhi iyini ezithakazelisayo kakhulu lesi sibalo cishe oyedwa iso, amehlo emibili nengxenye. Futhi kuye ngomongo (uma isithombe yomuntu ihembe ngamehlo zaveza), iso kungaba engaphezu kwemibili.

Ikhompyutha namanje akaqondi, kodwa "ubona"

Enye into eyenza kube nzima - kungcono ukukhanyisa. Isimo okufanayo ukukhanyisa ezahlukene uyobheka ahlukene. usayizi into ziyahlukahluka. Ngaphezu kwalokho, izinjongo noma isiphi isigaba. Ungasho kanjani ngendoda ukuthi ukuphakama kwakhe 2 amamitha? Lutho. ukukhula Human futhi kungase kube 2.3 m, futhi 80 cm. Nokho, njengoba ngezinye izinhlobo izinto, kukhona izinto of ekilasini efanayo.

Ikakhulukazi izinto eziphilayo ibhekane ezihlukahlukene nobunzima nezingcindezi. Izinwele abantu, abagijimi, izilwane. Bheka izithombe amahhashi egijima, ukunquma ukuthi kwenzekani nge mane zabo futhi umsila akunakwenzeka. A izinto agqagqene e isithombe? Uma Abasunduzani umfanekiso ikhompyutha, noma umshini enamandla kakhulu ukuthola nobunzima ukunikeza isinqumo esifanele.

buka Landelayo - kungcono esingalindelekile. Ezinye izinto, izilwane beziguqula imvelo, futhi ngempela ngobuciko. Futhi izindawo ezifanayo kanye Umbala. Noma kunjalo, sibabona, nakuba ngaso sonke isikhathi ekude.

Enye inkinga - ukunyakaza. Izinto in motion zicatshangwe ibhekane ukusonteka.

Abaningi izinto ziyahlukahluka kakhulu. Lapha, ngokwesibonelo, izithombe ezimbili ngezansi izinjongo "kusho uSihlalo".

Futhi kule uhlale. Kodwa ukufundisa umshini, yokuthi izinto ezihlukene ijamo, umbala, impahla, konke uyinto "kusho uSihlalo" - kunzima kakhulu. Lena yiyona nselelo. Ukuhlanganisa izindlela umbono ikhompyutha - ukufundisa umshini ukuqonda, ukuhlaziya, bayaqagela.

Ukudidiyelwa umbono ikhompyutha ezisekelweni ezahlukene

Sisindvo umbono ikhompyutha waqala ukungena ngaphezulu 2001, lapho edala ubuso umtshina kuqala. Senza ke abalobi ababili: Viola, Jones. Kwakuwusuku kuyashesha futhi kuthembekile algorithm kuqala ngokwanele, okuyinto ebonisa amandla izindlela zokufunda umshini.

Manje umbono ikhompyutha anisuthi zokusebenza ezintsha esisebenzayo - ukubonelelwa ubuso bomuntu.

Kodwa ukuqaphela nomuntu njengaku-movie - at engele okungahleliwe, izimo ezahlukene ukukhanyisa - akunakwenzeka. Kodwa ukuxazulula inkinga, noma eyodwa abantu ezahlukene ukukhanyisa ezahlukene noma ukuma ezahlukene, efanayo njengoba esithombeni ku iphasipoti, kungenzeka ngezinga eliphezulu ukuzethemba.

a izidingo ipasipoti isithombe kubangelwe isici ukuqashelwa ubuso algorithm.

Ngokwesibonelo, uma unayo i-passport yebhayomethrikhi, kwezinye zezindiza yesimanje, ungasebenzisa okuzenzakalelayo ipasipoti elawula uhlelo.

inkinga ezingaxazululeki umbono ikhompyutha - ikhono aqaphele yimuphi umbhalo

Mhlawumbe othile esetshenziswa OCR uhlelo. Enye yalezi - Reader lesidze, zithandwa kakhulu RuNet uhlelo. Kukhona izinhlobo eziningi lapho ugcwalisa idatha, basuke ukuskenwa kahle, ulwazi obonwa uhlelo kahle. Kodwa nge yimuphi umbhalo esithombeni ukuthi isimo sibi kakhulu. Le nkinga namanje uhlala ezingaxazululeki.

Imidlalo ezihilela umbono ikhompyutha, ezinyakazayo bamba

Hlukanisa endaweni enkulu - ukudalwa onobuhle ngakuthathu ezinyakazayo bamba (okuyinto ngempela ngempumelelo ukusebenza ngo imidlalo ye-computer). Uhlelo lokuqala, esebenzisa umbono ikhompyutha - uhlelo nokuxhumana nge-computer kusetshenziswa ukuthintwa. Lapho-ke sidalwe kwaba izinto eziningi evulekile.

I-algorithm yakhelwe impela nje, kodwa ukuze ulungiselele kwathatha ukudala generator izithombe zokwenziwa zabantu ukuba zithole kwesigidi izithombe. Supercomputer nabo ukukhetha kwemigomo algorithm, ngoba ngalo manje isebenza kahle.

Yilokho kwesigidi izithombe kanye ngesonto supercomputer letibalekako isikhathi kungenzeka ukudala algorithm ibusa 12% we umthamo owodwa processor ivumela umuntu ukuqonda isikhundla ngesikhathi sangempela. Lolu hlelo Microsoft Kinect (2010).

Ukucinga izithombe nokuqukethwe ikuvumela ukuba ulayishe izithombe ohlelweni, futhi imiphumela kuyokunikeza zonke izithombe ne okuqukethwe efanayo futhi wenza kusuka engela efanayo.

Izibonelo umbono kukhompyutha: ngakuthathu amabalazwe mgudumbili manje enziwa ngayo. Amamephu navigation izimoto zide ngokusho DVR.

Kukhona database ne izigidigidi izithombe yemibhalo enemininingwane yezindawo. By ukulanda isithombe database, ungakwazi ukunquma senziwa lapho, ngisho nabanye umbono. Yiqiniso, uma nje endaweni iyathandwa ngokwanele ukuthi ngesinye isikhathi izivakashi futhi wenza eziningi izithombe endaweni Iye yaba khona.

amarobhothi ayingxenye yamaphupho yonke indawo

Robotics ngesikhathi samanje, kuyo yonke indawo, ngaphandle kwalo nganoma iyiphi indlela. Manje kukhona izimoto ukuthi babe amakhamera akhethekile ukuthi aqaphele abahamba ngezinyawo kanye nezimpawu zomgwaqo ukudlulisa imiyalo umshayeli (lokhu ngendlela uhlelo lwekhompyutha ukubuka, kusiza umshayeli). Futhi kukhona kuyazenzakela zimoto, kodwa azikwazi bancike ohlelweni ikhamera yevidiyo ngaphandle kokusebenzisa semali ulwazi olwengeziwe.

ikhamera Modern - lokhu kuyinto obscura Analogue ikhamera

Ake sixoxe isithombe sedijithali. amakhamera Modern digital ahlelwe ku isimiso obscura ikhamera. Kuphela esikhundleni emgodini lapho ukukhanya kungena ugongolo futhi afakwa phezu kodonga ngemuva ekamelweni circuit kuncike, sinalo uhlelo olukhethekile optical ngokuthi lens. into yalo iwukuba ukuqoqa ugodo esikhulu ukukhanya bese siguqulwa ukuze wonke imisebe idlule iphuzu virtual ukuze bathole projector yakha isithombe ku ifilimu noma i-matrix.

Modern khamera (-matrix) yakhiwa izakhi ngabanye - pixels. kwe-pixel ngalinye lingaba ukukala amandla okukhanya okuyinto ngesenzakalo Imininingwane amaphikiseli, futhi ukukhipha inombolo eyodwa okukhipha. Ngakho-ke, ngo-ikhamera yedijithali, sithola esikhundleni isithombe ukugqama isethi ukukhanya izilinganiso, sabanjwa ngephikseli olulodwa - computer ensimini umbono. Ngakho-ke, lapho isithombe sibona hhayi ogeleza imigqa wabuka ecacile, futhi grid izikwele ezinemibala ngemibala ehlukene - pixels.

Ngezansi ubona isithombe sokuqala digital emhlabeni.

Kodwa lesi sithombe akuyona? Umbala. Kuyini umbala?

Umbono kwenqondo umbala

Umbala - yilokhu esikubonayo. Umbala owodwa futhi into efanayo ngoba abantu namakati zizohluka. Njengoba thina (abantu) futhi isilwane uhlelo optical - umbono ohlukile. Ngakho-ke, umbala - kungcono izinga ngokwengqondo umbono wethu ukuthi kwenzeka lapho ebona izinto nokukhanya. Futhi hhayi impahla ngokomzimba into futhi ukukhanya. Umbala - kuwumphumela nokuxhumana izingxenye ukukhanya, futhi isimo saleli isimiso sethu sokubona.

UHlelo Computer Vision in Python usebenzisa imitapo

Uma unqume ukuthi bahlanganyele sina ekuhloleni umbono ikhompyutha, kufanele ngokushesha ulungiselela eziningana nobunzima, lokhu isayensi akusiyo elula futhi ifihla inombolo izingibe. Kodwa "uHlelo Computer Vision phezu Python" zabhalwa-Jan-Erik Solema - incwadi esimayelana zonke ulimi kakhulu elula. Lapha uzothola bajwayelane izindlela kokubonwa lezinto ezihlukahlukene ku-3D, funda ukusebenza stereo isithombe, ngokoqobo virtual nezinye izinhlelo zokusebenza eziningi umbono ikhompyutha. Encwadini izibonelo okwanele Python. Kodwa izincazelo ethulwa, ngomqondo ongokomfanekiso, generalized, ukuze singathwesi overload ucwaningo kakhulu kanye nedatha kanzima. Sebenza efanelekayo abafundi, nayizimfundamakhwela, futhi abathanda. Download le ncwadi nabanye mayelana umbono ikhompyutha (PDF-format) kungaba inethiwekhi.

Okwamanje, kunezinhlobo umtapo yomthombo ovulekile umbono computer ubuchule obuphezulu kanye processing image futhi algorithm zezinombolo OpenCV. It is ukusebenza yesimanje izilimi eziningi zokuhlela, umthombo ovulekile. Uma sikhuluma umbono computer, i-Python isebenzisa njengoba ulimi lokwakha izinhlelo, kubuye ine ukusekelwa umtapo, e kwalokho, elokhu ivela futhi umphakathi omkhulu.

Le nkampani "Microsoft" inikeza amasevisi zayo Api-ngakwazi ukuqeqesha kwenethiwekhi ukusebenza ngezithombe zabantu. Kukhona ithuba ukusebenzisa umbono computer, i-Python isebenzisa njengoba ulimi lokwakha izinhlelo.