Amamodeli avamile amayunithi wokucubungula isithombe sekhamera eshisayo

Iyunithi yokucubungula isithombe yesithombe esishisayo ikhamera kuyisihluthulelo sokuguqula isignali yedijithali, ecutshungulwe isignali, ibe isithombe esishisayo esinembile. Amamodeli ajwayelekile enza kahle kakhulu emikhakheni ehlukahlukene.

I-ingka Technology's LC221 iyimojuli yekhamera ye-infrared esebenza kahle kakhulu. Isebenzisa izixazululo ezithuthukisiwe ze-ASIC, ezibhangqwe nomtshina we-infrared we-infrared we-infrared long-wave vanadium, enikeza izinqumo eziphezulu ezingu-384×288 noma 640×512. Iyunithi yayo enamandla yokucubungula izithombe isekela ukulungiswa kwe-NUC kususe ukukhanya okungalingani kwesithombe, kanye nama-algorithms ahlukahlukene njengokususwa kwe-pixel embi kanye nokunciphisa umsindo wesikhashana kanye nendawo ukuze kuthuthukiswe ikhwalithi yesithombe. Ukucutshungulwa kombala mbumbulu kungakwazi ukusatshalaliswa kwezinga lokushisa ngendlela enembile. Ukwengeza, isekela umbhalo wevidiyo we-H264/H265, ikwenze kube lula ukugcinwa nokudluliswa. Ivamise ukusetshenziswa ezimeni ezifana nosesho lwezindawo ezishisayo kanye nokuhlosisa okushisayo, njengokuthola abangeneleli ekuqapheni ukuphepha nokubikezela ukwehluleka kwemishini yezimboni.

Amayunithi ezithombe zemikhiqizo ye-FLIR ahlanganisa ama-chips okucubungula amasignali edijithali athuthukile, anikeza isivinini sokucubungula ngokushesha. Ngokwandisa ububanzi obuguqukayo bescale esimpunga ngokulinganisa kwe-histogram, umehluko ocashile wezinga lokushisa uyabonakala. Imodeli yokulungiswa kwezinga eliphezulu enembayo ephezulu iqinisekisa ukukalwa kwezinga lokushisa okunembile kanye nokukhanya okungaguquki kanye nombala endaweni yokushisa efanayo yesithombe. Iphinde isekele ukuhlanganisa, ukuhlanganisa izithombe ezishisayo nezithombe ezikhanyayo ezibonakalayo, ezisetshenziswa kakhulu ezimbonini ezidinga ukunemba okuphezulu ekuhlaziyeni izithombe, njengokuhlola kwezimboni kanye nokugadwa kwamandla, ukuze kwahlulele ngokunembile isimo sezisetshenziswa.

Le yunithi yokucubungula isithombe isebenzisa i-parallel computing architecture kanye nama-algorithms athuthukisiwe, amakhono aqinile wokucubungula wesikhathi sangempela kanye nokuqedela inani elikhulu lokucutshungulwa kwedatha yesithombe esishisayo ngesikhathi esifushane. Ngokuhlanganisa ama-algorithms okufunda okujulile kanye nokusetshenziswa kwamanethiwekhi e-convolutional neural ingakwazi ukuhlonza ngokuzenzakalelayo izinto eziqondiwe futhi iqaphe okudidayo kwezinga lokushisa okusekelwe kuzici zokucabanga okushisayo. Ekuqapheni umlilo wehlathi, ingakwazi ukubona ngokushesha imithombo yomlilo kanye nentuthu; ekuvikelekeni, ingahlaziya ukuziphatha okungavamile namazinga okushisa omzimba ezisebenzi. Idlala indima ebalulekile endaweni enkulu, ezimweni zokuqapha zesikhathi eside ezifana nokuqapha ezokuphepha kanye nomlilo wamahlathi, ukuqinisekisa ukuphepha kwesifunda.

Lawa mamodeli avamile amayunithi okucubungula izithombe, kanye nezinzuzo zawo zezobuchwepheshe, akhuthaza ukusetshenziswa nokuthuthukiswa kobuchwepheshe bokuthwebula okushisayo emikhakheni efana nemboni yezokuvikela. Esikhathini esizayo, ngokuthuthuka kobuchwepheshe, amamodeli asebenza kahle kakhulu azovela, athuthukise ukusebenza kanye nenani lesicelo samakhamera we-imaging ashisayo.