Ukuqhekeka okuthe kratya ngaphambili - ukuphela kunye nokubuya - ukuphela kwe-algorithms ekrelekrele kwiinkqubo zokuhlola

1. Ngaphambili - Ukuphela kwe-algorithm

Ingaphambili - ukuphelaI-algorithms isebenza ngqo ngaphakathi kwiyunithi yekhamera, ihlala ihlazisaI-computer yekhompyuthaamandla. Ezi algorithms zijolise kuQAPHELA Idatha yeSensor, ngaloo ndlela kuncitshiswa i-bandwidth kunye nomthwalo weserse ngokwenza imisebenzi yangaphambili kwinqanaba lekhamera. Makhe siphonononge izinto eziphambili:

a. Ikhamera yeHardware kunye ne-Shrisor Direction

Iikhamera zokujonga kwanamhlanje zibandakanya iintlobo ezininzi zenzwa:

• IZIQULATHO ZE-PMOS (CMOS, CCD): Bamba idatha ebonakalayo (imifanekiso kunye neevidiyo) phantsi kweemeko ezahlukeneyo zokukhanya.
• I-infrared (ir) yezinzwa: Vumela ikhamera ukuba ifake ividiyo ngokukhanya okuphantsi okanye ubumnyama obugcweleyo.
• I-lidar kunye neenzwa ezinzulu: Misela imigama emide kwaye ufumane izinto kwindawo ye-3D, iluncedo ukwahlula phakathi kwezinto nemvelaphi kwindawo.
• I-microPhones: Ngamanye amaxesha idityaniswe ne-audio - Ahlaluts esekwe.

Aba bavo bathumela idatha ye-Raw kwiCandelo leNkqubo, apho i-algorithms efanaImifanekiso yangaphambi - Ukulungiswaziyasetyenziswa.

b. Imifanekiso yangaphambi - Ukuqhubela phambili kunye nokuncitshiswa kwengxolo

Ngaphambi kokufaka naluphi na uhlalutyo olunzima,Imifanekiso yangaphambi - UkulungiswaKubalulekile ukuphucula umgangatho weintanethi, ngakumbi phantsi kweemeko ezimbi zokukhanyisa okanye iindawo ezinengxaki:

• Ingqondo ye-algorithms: Susa ingxolo yenzwa, ngokuqhelekileyo usebenzisa iifilitha njengeIGaussian BlUr or Ayisiyo - Ingingqi yeNdawo yeNgcaciso.
• Umahluko kunye nohlengahlengiso oluqaqambileyo: Algorithms enjeUkulungelelanisa i-fartogramLungisa ukukhanya kunye nokwahlukanisa ukomeleza ukubonakala.
• Ukuchongwa komda: Ukuhlolwa komda (E.G.,Umqhubi we-SOEL, Ukuchongwa komda) Unokunceda ekucaciseni imida yento, ebaluleke kakhulu ekutyikingqelekayo.

c. Ukufunyanwa kwentshukumo kunye nokukhupha umva

Ukuchongwa kwentshukumoyenye yemisebenzi esisiseko eyenziwe ngaphambili - ukuphela kwe-algorithms. Ihlala isekwe kumgaqo wokuthelekisa iifreyimu ezilandelelanayo zokufumana izinto ezihambayo.

• Ukukhutshwa kwemvelaphi: Ubuchule apho i-algorithm ikhupha imodeli yemvelaphi yereferensi ukusuka kwisakhelo sangoku. Naluphi na utshintsho olubalulekileyo lwenziwe luthando.
• Isakhelo sangahlukangaIndlela elula apho i-algorithm ibonisa umahluko phakathi kwezantya ezilandelelanayo, imimandla evuthayo apho utshintsho lwenzekile.
• Ukuhamba kwexesha: Indlela enobunkunkqele ehlalutya intshukumo ye-pixel kumacala alandelelanayo ekufumaneni intshukumo, ihlala isetyenziswa ngokudibeneyoI-Kalman Filtersngokulandela umkhondo.

d. Ukufunyanwa kwento kunye nokulandelela

Kumphambili - ukuphela, ukutshitshiswa kwento kunye nokulandelela kwenziwa ekuhlaleni ukukhangela kunye nokulandelela izinto (E.G., abantu, izilwanyana). Iindlela eziphambili zibandakanya:

• YOLO (jonga kanye)Ilizwe: le - I-Argorithm ye-Algorithm enokufumana izinto ezininzi ngokwenyani - Ixesha. I-Yolo yahlulahlula umfanekiso kwigridi kunye neebhokisi ezibonisa iibhokisi zento nganye kwigridi.
• I-Hoaar Cascade Classifiers: Isetyenziselwa imisebenzi yokufumana into elula, njengokuchongwa kobuso, ngokusekwe kwi-pre - iClassifiers.
• I-Kalman Gilter: Isetyenziselwaukulandela umkhondoIzinto ezihambahamba kwiifreyimu. Iqikelela imeko yento eshukumayo (isikhundla, i-velocity) kwaye icacisa imeko yayo yexesha elizayo.

e. Ukuchongwa kwe-anomaly kunye ne-triggers

Ukuchongwa kwe-anomaly ngaphambili - ukuphela kweziqhelo kugxile ekuchongeni iminyhadala engaqhelekanga kwindawo yokutya kwevidiyo:

• Ukuhamba ngequbulisoUkufumanisa iintshukumo ezikhawulezayo okanye ezingalindelekanga, ezinje ngomntu oqhuba okanye ngequbuliso.
• Umnqamlezo - Ukufunyanwa komgca: Sebenzisa ii-pieples windows okanye imigca ebaluleke kakhulu xa into iwela.
• Indawo yokungena: Kufumanisa ukuba into ingena okanye iphuma kwindawo echazwe kwangaphambili ngaphakathi kwesakhelo.

Ezi algorithms zinokuphelisa inyani - Ixesha lexeshaBuyela - IsipheloInkqubo okanye uthumele izaziso ezibonakalayo kubasebenzi bezokhuseleko.

2. Buyela - Ukuphunyezwa kwe-algorithm

IBuyela - IsipheloInkqubo inoxanduva lokuphakamisa okuphezulu, ukuphatha idatha yedatha entsokothi kwaye igcina imiqulu emikhulu yedatha yevidiyo. Isebenza ngokufumana imijelo yevidiyo okanye i-metadata ukusuka ngaphambili - iikhamera zokuphela kwaye zenza uhlalutyo lwe-Add, zihlala zisebenzisa iindlela zokufunda kunye noomatshini. Nantsi iqhekeza leimisebenzi ephambiliyenziwe emva - ukuphela kwe-algorithms:

a. Umjikelo weVidiyo kunye nokuHanjiswa kweDatha

• Ukuqokelelwa kwedatha: Iikhamera zidlulisela idatha yevidiyo
• Uxinzelelo: Ukunciphisa ukusetyenziswa kwebhendi, imijelo yevidiyo ihlala inyanzelekile ukuba isebenzise imigangatho efanaH.264 or H.265, egcina umgangatho wevidiyo ngelixa kuncitshiswa ubungakanani befayile.

b. Uhlalutyo lweVidiyo kunye nokufunda okunzulu

• UKUHLAZIYELWA: Umva - Ukuphela usebenzisa iimodeli zokufunda ezinzulu ngathiYolo, I-R - CNN, okanyeI-SSD. Ezi modeli ziqeqeshelwe iidatha ezinkulu ukuze zikwazi ukubona izinto ezahlukeneyo njengabantu, izithuthi, izilwanyana, njl.

• Ukwamkelwa kobuso: Ukungqinisisa isazisi okanye ukuhlolwa, i-algorithms yobuso isetyenziswa, ngokusekwe kwiimodyuli zokufunda ezinzulu ngathiI-PATSMT or Inzulu. Ezi modeli zithelekisa ubuso kwividiyo kwiziko ledatha yabantu abazaziwayo.

• Ukwamkelwa kwesenzoUkongeza ekuboneni izinto, umva - Ukuphela kunokwenza inyathelo okanye indlela yokuziphatha ngaphakathi kwividiyo. Umzekelo, ukufumanisa imilo, ukuhamba okukrokrelayo, okanye ezinye iindlela ezichazwe kwangaphambili usebenzisaI-RNNS (inethiwekhi ephindaphindiweyo) or Ii-3D ze-3D.

• Ukuhlelwa komsitho: Umva - Isiphelo senza izinto ezifunyanisiweyo okanye indlela yokuziphatha kwiziganeko ezinentsingiselo (i.G.

c. Ukumakishwa kweMetadata kunye nokukhangela

• Ukumakisha: Isakhelo ngasinye okanye ividiyo yevidiyo iphawulwe nge-metadata efanelekileyo (E.G., ixesha, indawo, izinto ezichongiweyo, iziganeko).
• Isalathiso: Ividiyo kunye nedatha yomnyhadala zikhonjisiwe ukuvumela ukukhangela ngokufanelekileyo. Sebenzisa iitekhnoloji ezinjeI-elasticsearch, kuya kuba lula ukukhangela izixa ezikhulu zedatha yevidiyo esekwe kwiithegi okanye kwi-metadata.

Umzekelo, unokukhangela "abantu abafunyenwe kwindawo ethintelweyo ukusuka kwi-2 pm ukuya kwi-3 pm."

d. Uhlalutyo lokuziphatha kunye nokufunyanwa kwe-anomaly

• Imeko yetekisi: Sebenzisa iimodeli zokufunda zoomatshini, inkqubo ifunda ukusuka kwizixa ezikhulu zedatha yembali zeziphi izinto eziqhelekileyo zokuziphatha ezikwindawo ezithile ezikwimo ngqo (E.G., ivenkile, kwikona yesitrato). Imodeli ke iflegi isuka kwisiqhelo.

• Unxibelelwano lomnyhadala: Buyela - Iinkqubo zokuphelisa zinokunxibelelana neziganeko ezininzi okanye imijelo yedatha (E.G., ukudibanisaUkuchongwa kwentshukumongeukwamkelwa kobuso). Ukuba umsebenzi ongaqhelekanga ufunyenwe, inkqubo inokuvelisa izilumkiso zentshukumo.

• Ixesha elide - Uhlalutyo lwexeshaNgexesha lokulandela umkhondo, le nkqubo inokulandelela iindlela kunye neepateni, zibonelela ngezakhono zobuqikelelo (E.G., zichonga iindawo ezinokubakho zobusela, ukucacisa xa imimandla ethile inokufumana i-gorge emsebenzini).

e. Ukudityaniswa kwelifu kunye nokunganyamezelani

• Ukugcinwa kwelifu: Idatha yevidiyo, ngakumbi i-up: ividiyo yenkcazo, inokugcinwa kwilifu, ivumela ukugcinwa okukhulu ngaphandle kokulayisha iziseko zophuhliso.

• I-AI yelifu AI: Abanye kwenziwa kwilifu ukuze basebenzise izixhobo zokusebenza ezingamandla (E.G.G., GPUS kwimisebenzi yokufunda enzulu). Ilifu linokusetyenziselwa ukuqeqesha iimodeli kwiidatha ezinkulu.

3. Imeko yesicelo

Ngobuchule obuphambili obuphambili be - Ukuphela kunye nokubuya - ukuphela kwe-algorithms ekrelekrele, iinkqubo zokujonga ngoku zisetyenziswa kwizicelo ezahlukeneyo:

a. I-Urbar Survese kwizixeko ezimkayo

• Ukuhlolwa kwetrafikhi: Iikhamera zinokujonga ukuhamba kwezithuthi, ziqwalasele iingozi, kunye nokulandelela izithuthi zokunyhashwa ngokutsha okanye ukuqhuba izibane ezibomvu.

• Ulawulo lweXesi: Iikhamera ezixhotyiswe ngabantu ukubala kunye nohlalutyo lokuziphatha ngendlela yokuziphatha

• Ukhuseleko loLuntu: Iikhamera zinokufumana indlela yokuziphatha engaqhelekanga (E.G., ukulwa okanye ukuThintelwa) kunye nabasemagunyeni ngoko nangoko.

b. Ukuhlolwa kwentengo yokuthengiswa kothintelo kunye nokuqonda kwabathengi

• Ukuthintela ubusela: I-Ai Algorithms ifumana indlela yokuziphatha erhanelayo enjengokuthengisa okanye ukungaqhelekanga kwiipateni zevenkile.

• I-Alutetics zabaThengi: Abathengisi banokusebenzisa iikhamera ukulandelela ukuhamba kwabathengi, hlalutya ukuba abathengi bachitha njani kumacandelo athile, kwaye basebenzele iindlela zokugcina izinto ngokubhekisele kwiipateni zendlela.

c. Ukhuseleko lwempilo kunye nesibhedlele

• Ukubekwa kweliso kwisigulana: Kwizibhedlele, iikhamera ezikrelekrele zinokubeka esweni iintshukumo zesigulana ukuze zikwazi ukubona, ukufikelela okungagunyaziswanga kwiindawo ezibuthathaka, okanye izigulana ekubandezelweni.

• UKhuseleko lwaBasebenzi: Abasebenzi bezokhuseleko banokufumana izilumkiso kwimeko yokuziphatha kakuhle okanye ukufikelela kwabasebenzi abangagunyaziswanga.

d. Ukukhusela izibonelelo ezibalulekileyo

• Iphezulu - Iindawo zokhuseleko: Iinkqubo zokuhlola zikhusela phezulu - Iindawo ezinamaxabiso anje ngamaziko edatha, izityalo zamandla kunye nezakhiwo zikarhulumente, apho isetyenziselwa khona i-algorithms yolawulo, ukuvunywa kobuso, kunye nokuchongwa kobuso.

e. Ukhuseleko lwasekhaya

• Ukuchongwa kwentshukumo: Kunqanyuzo lwasekhaya, iikhamera ezinomxholo wobuso kunye nokulandela umkhondo we-algorithms kunokuchonga abangeneleli, abanini bamakhaya abophayo, kunye neealamu ze-trigger.

• Iphakheji yokuthintela ubusela: Iikhamera zinokufumana imisebenzi erhanelayo enxulumene nokuthengiswa kwephakheji kwaye yazisa abanini-laya.

Ukuqukumbela

Ukudityaniswa kweI-algorithms ekrelekreleZombini ingaphambili - ukuphelakwayeBuyela - Isipheloiguqula intsimi yeukuhlolwa. Ukusuka ekufumaneni kwedatha yokuqala kunye nokufunyanwa kwesisiseko kwinqanaba lekhamera ukuya kwinqanaba lohlalutyo kunye nomatshini wokufunda kwiserver kwiserver kwiserver kunye ne-AlgorithMs, ezi algorith zibonelela ngezisombululo ezibanzi zamashishini ahlukeneyo. Njengoko ukufunda kunye nomatshini kuyaqhubeka nokuvela, ezi nkqubo ziya kuba namandla ngakumbi, ukubonelela ngokhuseleko oluphuculweyo, ulawulo olungcono lwezixhobo, kunye nobuchule obuqikethiweyo obunokuthintela izoyikiso ngaphambi kokuba zikhule.