Umtapo wolwazi wamahhala i-OpenCV 4.7 (I-Open Source Computer Vision Library) yakhululwa, ihlinzeka ngamathuluzi okucubungula nokuhlaziya okuqukethwe kwesithombe. I-OpenCV ihlinzeka ngama-algorithms angaphezu kuka-2500, kokubili akudala futhi abonisa intuthuko yakamuva ekuboneni ngekhompyutha nezinhlelo zokufunda zomshini. Ikhodi yomtapo wolwazi ibhalwe ngo-C++ futhi isatshalaliswa ngaphansi kwelayisensi ye-BSD. Izibopho zilungiselelwa izilimi ezihlukahlukene zokuhlela, kuhlanganise nePython, MATLAB neJava.
Umtapo wolwazi ungasetshenziswa ukubona izinto ezithombeni nakumavidiyo (isibonelo, ukubonwa kobuso nezibalo zabantu, umbhalo, njll.), ukulandelela ukunyakaza kwezinto namakhamera, ukuhlukanisa izenzo kuvidiyo, ukuguqula izithombe, ukukhipha amamodeli e-3D, ukukhiqiza isikhala se-3D kusuka ezithombeni ezisuka kumakhamera we-stereo, ukudala izithombe zekhwalithi ephezulu ngokuhlanganisa izithombe zekhwalithi ephansi, ukucinga izinto ezisesithombeni ezifana nesethi yezakhi eziphrezentwe, ukusebenzisa izindlela zokufunda umshini, ukubeka omaka, ukuhlonza izakhi ezivamile ezinhlobonhlobo ezihlukahlukene. izithombe, ezisusa ngokuzenzakalelayo iziphambeko ezifana neso elibomvu .
Phakathi kwezinguquko ekukhishweni okusha:
- Ukuthuthukiswa okubalulekile kokusebenza kwe-convolution kumojuli ye-DNN (Deep Neural Network) kwenziwe ngokusetshenziswa kwama-algorithms okufunda komshini ngokusekelwe kumanethiwekhi emizwa. I-algorithm ye-Winograd ye-convolution esheshayo isetshenzisiwe. Kwengezwe izendlalelo ezintsha ze-ONNX (Open Neural Network Exchange): I-Scatter, ScatterND, Tile, ReduceL1 kanye ne-ReduceMin. Ukwesekwa okwengeziwe kohlaka lwe-OpenVino 2022.1 kanye ne-CANN backend.
- Ikhwalithi ethuthukisiwe yokutholwa kwekhodi ye-QR nokukhishwa kwamakhodi.
- Kungezwe usekelo lomaka okubukwayo i-ArUco ne-AprilTag.
- Kungezwe i-tracker ye-Nanotrack v2 esekelwe kumanethiwekhi we-neural.
- Kusetshenziswe i-algorithm yokufiphalisa i-Stackblur.
- Usekelo olungeziwe lwe-FFmpeg 5.x ne-CUDA 12.0.
- Kuphakanyiswe i-API entsha ukuze kusetshenziswe amafomethi ezithombe ezinamakhasi amaningi.
- Kungezwe usekelo lwelabhulali ye-libSPNG yefomethi ye-PNG.
- I-libJPEG-Turbo inika amandla ukusheshisa kusetshenziswa imiyalo ye-SIMD.
- Kuplathifomu ye-Android, usekelo lwe-H264/H265 selusetshenzisiwe.
- Wonke ama-API ayisisekelo e-Python anikeziwe.
- Kwengezwe i-backend entsha yendawo yonke yemiyalo ye-vector.
Source: opennet.ru