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

1. Ngaphambili - Ukuphela kwe-algorithm

Ngaphambili - Isiphelo se-algorithms isebenza ngqo kwiyunithi yekhamera, idla ngokufumana igalelo lekhompyuter. Ezi algorithms zijolise ekuqhubekekeni idatha ye-RAW Sorson, ngaloo ndlela kuncitshiswa umthwalo we-bandwidth kunye nomthwalo weserver ngokwenza imisebenzi yangaphambili kwinqanaba lekhamera. Makhe siphonononge izinto eziphambili:

a. Ikhamera yeHardware kunye ne-Shrisor Direction

Iikhamera zokujonga kwanamhlanje zibandakanya iintlobo ezininzi zenzwa:

• Izinzwa zemifanekiso (i-CMOs, i-CCD): Faka idatha yeDatha (imifanekiso kunye neevidiyo) phantsi kweemeko ezahlukeneyo zokukhanyisa.
• I-infrared (ir) yezinzwa: Yenza ikhamera ikhuphe ividiyo ngesibane esisezantsi okanye ubumnyama obugcweleyo.
• I-lidar kunye nezinzwa ezinzulu: Misela imigama kunye nezinto ezibonakalayo kwindawo ye-3D, iluncedo ngokwahlula phakathi kwezinto kunye nemvelaphi kwindawo nemvelaphi.
• I-microPhones: ngamanye amaxesha ezidityaniswe kwi-audio - uhlaziyo olusekwe.

Ezi zinzwa zithumela idatha eluhlaza kwicandelo lenkqubo, apho algorithms njengomfanekiso wangaphambi - Ukusebenza kuyasetyenziswa.

b. Imifanekiso yangaphambi - Ukuqhubela phambili kunye nokuncitshiswa kwengxolo

Ngaphambi kokufaka naluphi na uhlalutyo olunzima, umfanekiso wangaphambi - Ukusebenza kubalulekile ukuphucula umgangatho weintanethi, ngakumbi phantsi kweemeko ezimbi zokukhanyisa okanye iindawo ezinomsindo:

• Ukuchonga i-algorithms: Susa ingxolo yenzwa, ngokuqhelekileyo usebenzisa iifilitha njenge-Gaussian Blur okanye i-no - indawo yendawo yokubonisa.
• Ukuthelekisa kunye nohlengahlengiso oluqaqambileyo: I-algorithms efana ne-Alaptiber Hiatogram ihlengahlengisa ukukhanya kunye nokwahlukanisa ukuphucula ukubonakala.
• Ukuchongwa komda: Ukuchongwa komda

c. Ukufunyanwa kwentshukumo kunye nokukhupha umva

Ukufunyanwa kwentshukumo yenye yemisebenzi esisiseko eyenziweyo ngaphambili - ukuphela kwe-algorithms. Ihlala isekwe kumgaqo wokuthelekisa iifreyimu ezilandelelanayo zokufumana izinto ezihambayo.

• Ukuthatyathwa kwemvelaphi: ubuchule apho i-algorithm ikhupha imodeli yemvelaphi yereferensi ukusuka kwisakhelo sangoku. Naluphi na utshintsho olubalulekileyo lwenziwe luthando.
• Isakhelo sahluka-hlula: indlela elula apho i-algorithm ibonisa umahluko phakathi kwezantya ezilandelelanayo, imimandla evuthayo apho utshintsho lwenzekile.
• Ukuhamba kweenyawo: Eyona ndlela inobunkunkqele ehlalutya intshukumo ye-pixel kumacala athile alandelelanayo ekufumaneni intshukumo, edla ngokusetyenziswa kunye neefilitha ze-kalman zokulandela 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:

• I-Yolo (ujonga kuphela I-Yolo yahlulahlula umfanekiso kwigridi kunye neebhokisi ezibonisa iibhokisi zento nganye kwigridi.
• I-Hoare Classimiers: isetyenziselwa imisebenzi yokufumana into elula, njengokuchongwa kobuso, ngokusekwe kwi-pre - iClassifiers.
• Icebo lokucoca i-Kalman: esetyenziselwa ukulandela umkhondo izinto ezihambayo. 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 ngequbuliso: Ukufunyanwa kweentshukumo ezikhawulezayo okanye ezingalindelekanga, ezinje ngomntu oqhuba okanye ngequbuliso.
• Umnqamlezo - Ukufunyanwa komgca: Sebenzisa ii-riberes ezibonakalayo okanye imigca ebaluleke kakhulu xa into iwela.
• I-Insusion yendawo: Vula ukuba into ingena okanye iphuma kwindawo echazwe kwangaphambili ngaphakathi kwesakhelo.

Ezi algorithms zinokuphelisa inyani - I-SUD IMIQATHANGO YAMVUZO - Isiphelo senkqubo okanye sithumele izaziso ezibonakalayo kubasebenzi bezokhuseleko.

2. Buyela - Ukuphunyezwa kwe-algorithm

Umva - Inkqubo yokuphela inoxanduva lokuphakamisa okuphezulu, ukuphatha idatha yedatha entsokothi kwaye igcina imiqulu emikhulu yedatha. Isebenza ngokufumana imijelo yevidiyo okanye i-metadata ukusuka ngaphambili - iikhamera zokuphela kwaye zenza uhlalutyo lwe-Add, zihlala zisebenzisa iindlela zokufunda kunye noomatshini. Nantsi iqhekeza lemisebenzi ephambili eyenziweyo nge-Mumva - ukuphela kwe-algorithms:

a. Umjikelo weVidiyo kunye nokuHanjiswa kweDatha

• Ukuqokelelwa kwedatha: Iikhamera zidlulisela idatha yevidiyo
• Uxinzelelo: Ukunciphisa ukusetyenziswa kwebhendi, imilambo yevidiyo ihlala inyanzelekile ukuba isebenzise imigangatho efana ne-H.264 okanye i-H.265, i Gcina ubungakanani bevidiyo ngelixa kugcinwe ubungakanani bevidiyo ngelixa ugcina ubungakanani bevidiyo ngelixa ubeka ubungakanani befayile.

b. Uhlalutyo lweVidiyo kunye nokufunda okunzulu

• Ukuchongwa kwento Ezi modeli ziqeqeshelwe iidatha ezinkulu ukuze zikwazi ukubona izinto ezahlukeneyo njengabantu, izithuthi, izilwanyana, njl.

• Ukuqwalaselwa kobuso Ezi modeli zithelekisa ubuso kwividiyo kwiziko ledatha yabantu abazaziwayo.

• Ukuqwalaselwa kwesenzo: Ukongeza ekuboneni izinto, umva - Ukuphela kunokwenza inyathelo okanye indlela yokuziphatha ngaphakathi kwividiyo. Umzekelo, ukufumanisa umlo, ukuhamba okukrokrelayo, okanye ezinye iindlela zokuziphatha ezichazwe kwangaphambili zisebenzisa i-RNNS (iinethiwekhi eziphindaphindiweyo) okanye ii-3D ze-CNS.

• Ukuhlelwa komnyhadala: Umva - Ukuphela kuhlelwa izinto ezichongiweyo okanye indlela yokuziphatha kwiziganeko ezinentsingiselo (i.G., "Umntu ochongiweyo", "izithuthi zenza inqwelo moya").

c. Ukumakishwa kweMetadata kunye nokukhangela

• Ukumakisha: Isakhelo ngasinye okanye i-upod yevidiyo iphawulwe nge-metadata efanelekileyo (E.G., ixesha, izinto, izinto ezichongiweyo, imicimbi).
• Isalathiso: Ividiyo kunye nedatha yomnyhadala zikhonjisiwe ukuvumela ukufuna ngokufanelekileyo. Sebenzisa iitekhnoloji ezinjenge-elastictor, 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

• I-POSTETIONOTIONT: Sebenzisa iimodeli zokufunda ngoomatshini, inkqubo ifunda ukusuka kwizixa ezikhulu zedatha yembali zeziphi izinto eziqhelekileyo ezikwindawo ezithile ezikwimo ekhethekileyo (E.G., ivenkile yesitrato). Imodeli ke iflegi isuka kwisiqhelo.

• Ukulungelelaniswa komsitho: Buyela - Iinkqubo zokuphelisa zinokunxibelelana neziganeko ezininzi okanye imijelo yedatha (E.G., Ukudityaniswa kokufunyanwa kwentshukumo). Ukuba umsebenzi ongaqhelekanga ufunyenwe, inkqubo inokuvelisa izilumkiso zentshukumo.

• Ixesha elide - Uhlalutyo lwexesha: Ukuhamba kwexesha kunokulandelela imikhondo kunye neepateni, ukubonelela ngamandla okuqikelelwa (E.G., ukuchonga iindawo ezinokubakho xa imimandla ethile inamava entshukumo yomsebenzi).

e. Ukudityaniswa kwelifu kunye nokunganyamezelani

• Ukugcinwa kwelifu: Idatha yevidiyo, ngakumbi - Ividiyo ephezulu, inokugcinwa kwilifu, ivumela ukugcinwa okungqongqo ngaphandle kokulayisha iziseko zophuhliso.

• Ukulungisa i-AI yelifu: Ezinye iinkqubo zenziwa efini ukuze zisebenzise izixhobo ezinamandla (i.G.G., GPUS IMISEBENZI EYENZIWEYO). 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 kwezendlela: Iikhamera zinokujonga ukuhamba kwezithuthi, ziqwalasele iingozi, kunye nokulandelela izithuthi zokunyhashwa ngokutsha okanye ukuqhuba izibane ezibomvu.

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

• Ukhuseleko loLuntu: Iikhamera zinokukufumanisa isimilo esingaqhelekanga (E.G., ukulwa okanye ukuThintelwa) kunye nabasemagunyeni ngoko nangoko.

b. Ukuhlolwa kwentengo yokuthengiswa kothintelo kunye nokuqonda kwabathengi

• Uthintelo loBubi: I-Ai Algorithms ifumana indlela yokuziphatha enyanzelayo njengeendlela zokuthengisa okanye ezingaqhelekanga kwintshukumo yevenkile.

• Uhlalutyo lwabathengi

c. Ukhuseleko lwempilo kunye nesibhedlele

• Ukubek'esweni kwesigulana: Kwizibhedlele, iikhamera ezikrelekrele kunokubeka iliso kwiintshukumo zesigulana ukuze zikwazi ukubona ukuwa, ukufikelela okungagunyaziswanga kwiindawo ezibuthathaka, okanye izigulana ekubandezelweni.

• Ukhuseleko lwabasebenzi: Abasebenzi bezokhuseleko banokufumana izilumkiso kwimeko yokuziphatha kakuhle okanye ukufikelela kwabasebenzi abangagunyaziswanga.

d. Ukukhusela izibonelelo ezibalulekileyo

• Iindawo eziphezulu

e. Ukhuseleko lwasekhaya

• Ukuchongwa kwe-Intrudududer: ekukhuleni kweKhaya, iikhamera ezihambelana nobuso kunye nokulandela umkhondo we-algorithms kunokuchonga abangeneleli, abanini bamakhaya abophayo, kunye neealarms ezixineneyo.

• Iphakheji yokuthintela ubusela: Iikhamera zinokufumana imisebenzi ekrokrisayo enxulumene nokuthengiswa kwephakheji kunye nokwazisa abanini bamakhaya.

Ukuqukumbela

Ukudityaniswa kwe-algorithms ekrelekrele kuyo bobabini ngaphambili - ukuphela kunye nokubuya - ukuphela kuyayitshintsha intsimi yokujonga. 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.