Ukuqhekeka okuningiliziwe ngaphambili - ukuphela nangemuva - qeda ama-algorithms ahlakaniphile ezinhlelweni zokubheka

1. Ophambili - Ukuphela kwe-algorithm

Le khasingaphambili - ukuphelaAma-algorithms asebenza ngqo ngaphakathi kweyunithi yekhamera, imvamisaI-Edge Computingamakhono. Lawa ama-algorithms ahlose ukukwenzacubungula idatha ye-sensor eluhlaza endaweni yangakini, ngaleyo ndlela enciphisa umthwalo we-bandwidth kanye neseva ngokwenza imisebenzi yokuqala ezingeni lekhamera. Ake sihlole izakhi eziphambili:

a. I-Hardware yekhamera nokuhlanganiswa kwenzwa

Amakhamera we-Modern Reviillance afaka izinhlobo eziningi zezinzwa:

• Izinzwa zezithombe (ama-CMO, CCD): Capture idatha ebonakalayo (izithombe namavidiyo) ngaphansi kwezimo zokukhanyisa ezihlukile.
• Izinzwa ezifakiwe (IR): Nika amandla ikhamera ukuthwebula ividiyo ekukhanyeni okuphansi noma ubumnyama obuphelele.
• Izinzwa ze-lidar kanye nokujula: Kala amabanga futhi ubone izinto endaweni ye-3D, ewusizo ekuhlukaniseni phakathi kwezinto nesizinda endaweni yesehlakalo.
• Amakrofoni: Kwesinye isikhathi kuhlanganiswe umsindo - Kususelwa ku-analytics.

Lezi zinzwa zithumela idatha eluhlaza kwiyunithi yokucubungula, lapho ama-algorithms afana khonaImage Pre - Iyacubungulaziyasetshenziswa.

b. Isithombe pre - ukucubungula kanye nokuncishiswa komsindo

Ngaphambi kokufaka noma yikuphi ukuhlaziya okuyinkimbinkimbi,Image Pre - Iyacubungulakubalulekile ukuthuthukisa ikhwalithi ye-footage, ikakhulukazi ngaphansi kwezimo ezingezinhle zokukhanyisa noma izindawo ezinomsindo:

• I-algorithms: Susa umsindo we-sensor, ngokujwayelekile usebenzisa izihlungi ezinjeIGaussian Blur or non - indawo kusho ukuphikisana.
• Qhathanisa nokulungiswa kokukhanya: Ama-algorithms afanaUkulinganiswa kwe-histogram okuguqukayoLungisa ukugqama futhi umehluko ukuthuthukisa ukubonakala.
• Ukutholwa komphetho: Ukutholwa komngcele (isb.,U-opharetha we-sobel, Ukutholwa komphetho kwe-cann) Kungasiza ukuchaza imingcele yento, ebaluleke kakhulu ekulandeleleni kwento.

c. Ukutholwa kokunyakaza kanye nokukhipha ngemuva

Ukutholwa kokutholwaUngomunye wemisebenzi eyisisekelo eyenziwe ngaphambili - ukuphela kwama-algorithms. Imvamisa isuselwa kumgomo wokuqhathanisa ozimele olandelanayo ukuthola izinto ezihambayo.

• Ukukhishwa kwangemuva: Indlela lapho i-algorithm isusa imodeli yangemuva ethenjwayo kusuka kufreyimu yamanje. Noma yiluphi ushintsho olukhulu luhlatshwe umkhosi njengokunyakaza.
• Umehluko wohlaka: Indlela elula lapho i-algorithm ikhombisa umehluko phakathi kwamafreyimu alandelanayo, i-Flackging izifunda lapho kwenzeka khona izinguquko.
• Ukugeleza kwe-Optical: Indlela eyinkimbinkimbi ngokwengeziwe ehlaziya ukunyakaza kwama-pixel ukuqina kwamafreyimu alandelanayo ukuthola ukunyakaza, okuvame ukusetshenziswa ngokubambisana nakhoIzihlungi zeKalmanngokulandela umkhondo.

d. Ukutholwa kwento nokulandelela

Ngaphambili - Ukuphela, ukutholwa kwento nokulandela umkhondo kwenziwa endaweni yakini ukukhomba nokulandela izinto (e.g., abantu, izimoto, izilwane). Amasu amakhulu afaka:

• Yolo (ubheka kanye kuphela): Isimo - se - The Art Algorithm angathola izinto eziningi ngokoqobo - isikhathi. I-Yolo ihlukanisa isithombe kwigridi futhi ibikezela amabhokisi okubopha entweni ngayinye kugridi.
• I-Haar Cascade Crisefiers: Isetshenziselwa ukuthola imisebenzi elula yokutholwa kwemisebenzi, njengokutholwa kobuso, okusekelwe ku-pre - ama-classififiers aqeqeshiwe.
• Isihlungi seKalman: KusetshenziselweukulandelelaUkuhambisa izinto ngaphesheya ozimele. Ilinganisa isimo sento eshukumisayo (isikhundla, velocity) futhi ibikezela isikhundla sayo esizayo.

e. Ukutholwa kwe-anomaly kanye nezimbangela zemicimbi

Ukutholwa kwe-anomaly ngaphambili - ngokujwayelekile kugxile ekuboneni imicimbi engajwayelekile ekuphathweni kwevidiyo:

• Ukunyakaza okungazelelwe: Ukutholwa kokunyakaza okusheshayo noma okungalindeleki, njengomuntu ogijimayo noma ukwakheka kwesixuku okungazelelwe.
• Isiphambano - Ukutholwa komugqa: Isebenzisa ama-trigwires atholakalayo noma imigqa edala izexwayiso lapho into ibabhuqa.
• Ukungena kwendawo: Thola uma into ingena noma iphuma endaweni echazwe ngaphakathi kohlaka.

Lawa ma-algorithms angabangela kwangempela - izexwayiso zesikhathi seemuva - ukuphelauhlelo noma thumela izaziso ezisheshayo kubasebenzi bezokuphepha.

2. Emuva - ukuphela kwe-algorithm ukusetshenziswa

Le khasiemuva - ukuphelaUhlelo lubhekene nokuphakamisa okusindayo, ukuphatha ama-analytics alunkimbinkimbi kanye nokugcina amavolumu amakhulu wedatha yevidiyo. Isebenza ngokuthola ukusakazwa kwevidiyo noma i-metadata kusuka ngaphambili - amakhamera okugcina futhi enze ukuhlaziywa okuthuthukile, imvamisa usebenzisa amasu wokufunda we-AI kanye nomshini. Nakhu ukuwohloka kweImisebenzi ebalulekilekwenziwe emuva - ukuphela kwama-algorithms:

a. Ukusakazwa kwevidiyo nokuhanjiswa kwedatha

• Ukuqoqwa kwedatha: Amakhamera adlulisa idatha yevidiyo emhlane - ukuphela kungaba ngokuxhumeka kwe-inthanethi okuqondile, amanethiwekhi wendawo (ama-lans), noma izinsizakalo zamafu.
• Ukucindezelwa: Ukwehlisa ukusetshenziswa kwe-bandwidth, ukusakazwa kwevidiyo kuvame ukucindezelwa kusetshenziswa amazinga afanaH.264 or H.265, egcina ikhwalithi yevidiyo ngenkathi inciphisa usayizi wefayela.

b. Ukuhlaziywa kwevidiyo nokufunda okujulile

• Ukutholwa Kwento: Umhlane - ukuphela kusetshenziswa amamodeli wokufunda ajulile afanaYolo, Ngokushesha r - cnn, nomaI-SSD(I-Multibox Detector Multibox) yokutholwa okunembile kakhulu kanye nokuhlukaniswa. Lawa mamodeli aqeqeshelwe kuma-datasets amakhulu ukubona izinto ezahlukahlukene ezinjengezinto, izimoto, izilwane, njll.

• Ukuqashelwa kobuso: Ukuqinisekiswa kobunikazi noma ukubhekwa, ukubonwa kobuso kwe-algorithms kusetshenziswa, ngokuvamile kususelwa kumamodeli wokufunda ajulile afanaFacenet or I-depApp. Lawa mamodeli aqhathanisa ubuso e-video bootage kuya ku-database yabantu abaziwayo.

• Ukuqashelwa Kwesenzo: Ngaphezu kokuthola izinto, emuva - ukuphela kungahlukanisa nezenzo noma ukuziphatha kuvidiyo. Isibonelo, ukuthola ukulwa, ukunyakaza okusolisayo, noma okunye ukuziphatha okuchaziwe kusetshenziswaAma-RNNS (amanethiwekhi we-neural aphindaphindayo) or 3d CNNS.

• Ukuhlukaniswa Kwemicimbi: Umhlane - Ukuphela kuzanda izinto ezitholakele noma ukuziphatha okutholakele emicimbini enenjongo (e.g., "umuntu etholakele", "imoto ipake inqobo kakhulu"

c. Ukumaka kwe-metadata nokusesha

• Umakoqa: Uhlaka ngalunye noma ingxenye yevidiyo isungulwe nge-metadata efanelekile (e.g., isikhathi, indawo, izinto ezihlonziwe, imicimbi).
• Okukhomba: Idatha yevidiyo neyomcimbi ikhonjiswa ukuvumela ukusesha okusebenzayo. Kusetshenziswa ubuchwepheshe obunjengeNanyastics, kuba lula ukusesha ngamanani amakhulu wedatha yevidiyo esekelwe kumathegi noma ku-metadata.

Isibonelo, ungasesha "abantu abatholwe endaweni ekhawulelwe kusuka ku-2 PM kuya ku-3 PM."

d. Ukuhlaziywa kokuziphatha nokutholwa kwe-anomaly

• Ukuqashelwa Kwephethini: Kusetshenziswa amamodeli wokufunda umshini, uhlelo lufunda kusuka kumanani amakhulu emininingwane emlandweni ukuthi yiziphi izindlela zokuziphatha ezikhethekile (e.g. Isitolo, ekhoneni lomgwaqo). Imodeli bese ibamba amafulegi liphambuka kusuka evamile.

• Ukuhlangana komcimbi: Emuva - Izinhlelo zokuphela zingaxhumanisa imicimbi eminingi noma imifudlana yedatha (isb., Ukuhlanganisaukutholwa kokutholwane-Ukuqashelwa kobuso). Uma umsebenzi ongajwayelekile utholakele, uhlelo lungakhiqiza izexwayiso ezisebenzisekayo.

• Kudala - Ukuhlaziywa Kwesikhashana: Ngokuhamba kwesikhathi, uhlelo lungakwazi ukulandelela izitayela namaphethini, lunikeza amandla okuqagela (isb., Ukuhlonza izindawo ezingaba khona zokweba, ukubikezela lapho izindawo ezithile zingathola khona ukuhlinzwa emsebenzini).

e. Ukuhlanganiswa kwamafu nokukhubazeka

• Isitoreji samafu: Idatha yevidiyo, ikakhulukazi ephakeme - Ukuchazwa kwevidiyo, kungagcinwa efwini, kuvumela isitoreji esilinganiselwe ngaphandle kokulayisha kakhulu ingqalasizinda yendawo.

• ICloud Ai Iyacubungula: Ukucutshungulwa okuthile kwenziwa efwini ukuze usebenzise ngokunenzuzo i-Hardware enamandla (E.G., GPUS yemisebenzi yokufunda ejulile). Ifu lingasetshenziswa futhi ukuqeqesha amamodeli kuma-dataset amakhulu.

3. Izimo zesicelo

Ngamakhono athuthukile ngaphambili - ukuphela nangemuva - ukuphela kwe-algorithms ehlakaniphile, amasistimu wokubheka manje asetshenziswa ezinhlelweni ezahlukahlukene:

a. Ukubhekwa kwedolobha emadolobheni ahlakaniphile

• Ukuqapha Kwethrafikhi: Amakhamera angabheka ukugeleza komgwaqo, atholwe izingozi, futhi alandelele izimoto zokwephulwa njengokushesha noma ukugijima amalambu abomvu.

• Ukuphathwa kwesixuku: Amakhamera ahlome abantu ababala kanye nokuhlaziywa kwama-algorithms asiza ukuphatha ukunyakaza kwesixuku, ukuqinisekisa ukuphepha ezindaweni zomphakathi.

• Ukuphepha Komphakathi: Amakhamera angabona ukusebenza okungajwayelekile (isb.g.

b. Ukubhekwa kokuthengisa kokuvimbela ukweba kanye nokuqonda kwamakhasimende

• Ukuvimbela Ukweba: I-AI algorithms ibona izindlela ezisolisayo ezifana namapulangwe noma amaphethini angajwayelekile ekunyakazeni kwe-shopper.

• Ukuhlaziywa kwamakhasimende: Abathengisi bangasebenzisa amakhamera ukulandelela ukuhamba kwamakhasimende, bahlaziye amakhasimende amade ukuthi achitha isikhathi esingakanani ezingxenyeni ezithile, futhi abeke ezakhiweni zezakhiwo zesitolo ezisuselwa kumaphethini omgwaqo.

c. Ukuphepha kwezempilo kanye nesibhedlela

• Ukuqapha Besiguli: Ezibhedlela, amakhamera ahlakaniphile wokuhlola angabheka ukunyakaza kweziguli ukuthola ukuwa, ukufinyelela okungagunyaziwe kwezindawo ezibucayi, noma ezigulini ezicindezelekile.

• Ukuphepha Abasebenzi: Abasebenzi bezokuphepha bangathola izexwayiso uma kunesihluku sokuziphatha noma ukufinyelela kwabasebenzi abangagunyaziwe.

d. Ukuvikelwa Kwengqalasizinda Ebucayi

• Ukuphakama - izindawo zokuphepha: Izinhlelo zokubhekwa zivikela ukuphakama - inani lezindawo ezifana nezikhungo zedatha, izitshalo zamandla, kanye nezakhiwo zikahulumeni, lapho kusetshenziselwa ama-algorithms ukuze uthole ukulawulwa kobuso, ukuqashelwa kobuso, nokutholwa kwe-anomaly.

e. Ezokuphepha Ekhaya

• Ukutholwa kwe-IntruderEzokuphepha Kwasekhaya, amakhamera anokuqashelwa kobuso kanye nama-algorithms alandelwayo atholakalayo angakhomba abangenayo, abaninikhaya abavuselelayo, kanye nama-alamu abangela.

• Iphakheji yokuvimbela ukwebiwa: Amakhamera angabona imisebenzi esolisayo ephathelene nokweba kwephakeji futhi wazise abanini khaya.

Ukugcina

Ukuhlanganiswa kweIntelligent algorithmsKokubilingaphambili - ukuphelana-emuva - ukuphelaukuguqula inkambu yeukugada. Ukusuka ekutholeni idatha yokuqala kanye nokutholwa komcimbi okuyisisekelo ezingeni lekhamera kuya kuma-analytics athuthukile kanye nokufunda komshini kwiseva - uhlangothi, la ma-algorithms ahlinzeka ngezixazululo eziphelele zemikhakha ehlukahlukene. Njengoba i-AI ne-Machine Learning qhubeka nokuvela, lezi zinhlelo zizoba namandla ngokwengeziwe, zinikeza ukuphepha okuthuthukisiwe, ukuphathwa kwezinsiza ezingcono, kanye namakhono okuqagela angavimba izinsongo ezingaba khona ngaphambi kokuba zikhule.