IMPLEMENTATION OF STEREO VISION USING GPU

V uvodnem delu diplomskega dela smo predstavili področje stereovida. Opisali smo potek rekonstrukcije globinskih podatkov iz stereo para slik ter možne pristope in metode, katere se uporabljajo za rekonstrukcijo okolja. V nadaljevanju smo predstavili centralne procesne enote, grafične procesne enote ter opisali glavne razlike med njimi. Predstavili smo programske vmesnike namenjene programiranju grafičnih enot in opisali knjižnico CUDA, katera se uporablja za programiranje grafičnih enot proizvajalca Nvidia. Iz programa namenjenega za izvajanje na CPU smo izdelali program za GPU in opravili primerjave hitrosti rekonstukcije. Za konec smo izdelali uporabniški vmesnik v programskem paketu LabView, in opravili preizkus rekonstrukcije v realnem času s pomočjo spletnih kamer.In the introductory part of the thesis we have presented the area of stereo vision. We described the course of the reconstruction of depth information from stereo pairs of images and possible approaches and methods which are used to reconstruct the environment. In the continuation we presented central processing units (CPU), graphic processing units (GPU) and described the main differences between them. We presented programme interfaces intended to program graphic units and described CUDA library, which is used to program graphic units of producer Nvidia. From the program designed to run on the CPU we have produced a program for GPU and we made comparison of the speed of reconstruction. For conclusion we developed the user interface in program packet LabView and passed the test of reconstruction in real time with the help of web cameras