Sakin ɗakin karatu na hangen nesa na kwamfuta OpenCV 4.2

ya faru sakin ɗakin karatu kyauta Buɗe CV 4.2 (Open Source Computer Vision Library), wanda ke ba da kayan aiki don sarrafawa da nazarin abubuwan hoto. OpenCV yana ba da algorithms sama da 2500, duka na al'ada da kuma nuna sabbin ci gaba a hangen nesa na kwamfuta da tsarin koyon injin. An rubuta lambar ɗakin karatu a C++ da rarraba ta ƙarƙashin lasisin BSD. Ana shirya ɗaure don harsunan shirye-shirye daban-daban, gami da Python, MATLAB da Java.

Ana iya amfani da ɗakin karatu don gane abubuwa a cikin hotuna da bidiyo (misali, fahimtar fuskoki da alkaluman mutane, rubutu, da sauransu), bin diddigin motsin abubuwa da kyamarori, rarraba ayyuka a cikin bidiyo, canza hotuna, fitar da samfuran 3D, samar da sararin samaniya na 3D daga hotuna daga kyamarori na sitiriyo, ƙirƙirar hotuna masu inganci ta hanyar haɗa ƙananan hotuna, neman abubuwa a cikin hoton da suka yi kama da abubuwan da aka gabatar, yin amfani da hanyoyin koyo na inji, sanya alamomi, gano abubuwan gama gari a cikin daban-daban. hotuna, ta atomatik kawar da lahani kamar ja-ido .

В sabo saki:

• An ƙara baya don amfani da CUDA zuwa tsarin DNN (Deep Neural Network) tare da aiwatar da algorithms koyo na inji dangane da hanyoyin sadarwar jijiya kuma an aiwatar da tallafin API na gwaji. nGraph OpenVINO;
• Yin amfani da umarnin SIMD, an inganta aikin lambar don fitowar sitiriyo (StereoBM/StereoSGBM), sakewa, masking, juyawa, lissafin abubuwan da aka rasa launi da sauran ayyuka masu yawa;
• Ƙara aiwatar da zaren da yawa na aikin pyrDown;
• Ƙara ikon fitar da rafukan bidiyo daga kwantena na kafofin watsa labaru (demuxing) ta amfani da backend videoio dangane da FFmpeg;
• Algorithm ɗin da aka ƙara don saurin sake zaɓen mitar hotuna da suka lalace FSR (Sake Gina Zaɓaɓɓen Sauti);
• Hanyar da aka ƙara RIC don haɗawa da wuraren da ba a cika ba;
• Ƙara hanyar daidaitawa Logos;
• Tsarin G-API (opencv_gapi), wanda ke aiki azaman injiniya don ingantacciyar sarrafa hoto ta amfani da algorithm na tushen jadawali, yana goyan bayan ƙarin hadaddun hangen nesa na kwamfuta da zurfin koyon injina. An bayar da goyan baya ga ingin Inference Engine backend. Ƙara goyon baya don sarrafa rafukan bidiyo zuwa tsarin aiwatarwa;
• An kawar rauni (CVE-2019-5063, CVE-2019-5064), wanda zai iya haifar da kisa ga masu hari lokacin aiwatar da bayanan da ba a tantance ba a cikin tsarin XML, YAML da JSON. Idan an ci karo da harafi mai lambar banza yayin nazarin JSON, ana kwafi gabaɗayan ƙimar zuwa ma'ajin, amma ba tare da bincika da kyau ba ko ya wuce iyakokin yankin da aka keɓe.

source: budenet.ru