Algoritam za procjenu pokreta pri kodiranju videa proširen informacijama o kretanju kamere

master's thesis

• Title (english): Video encoding motion estimation algorithm augmented with camera motion information
• Author: Marko Barbir
• Mentor: Daniel Hofman (mentor)
• Committee member: Daniel Hofman (član povjerenstva)
• Committee member: Mario Kovač (član povjerenstva)
• Committee member: Igor Čavrak (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-09-20, Croatia

Abstract

Bespilotne letjelice s pogledom iz prvog lica (eng. First Person View, FPV) korisne su u raznim primjenama. Velika prepreka kod FPV letjelica je kvaliteta i brzina kašnjenja videa s letjelice do pilota kako bi bolje mogao s njom upravljati. Ovaj rad bavi se iskorištavanjem informacije upravljačkih naredbi pilota kako bi se skratilo vrijeme trajanja jednog od zahtjevnijih dijelova algoritma HEVC kompresije videa - procjene pokreta, rezultirajući manjim kašnjenjem. Istraženo je razvijanje modela strojnog učenja koje koristi takve informacije za predviđanje vektora pokreta slike s referentnom, te je prikazano njegovo izvođenje unutar već postojećeg HEVC enkodera otvorenog koda.

Abstract (english)

Unmanned aerial vehicles with first-person view (FPV) are useful in various applications. A significant challenge with FPV drones is the quality and latency of the video from the vehicle to the pilot to facilitate better control. This work focuses on utilizing the information from the pilot’s control commands to reduce the duration of one of the more demanding parts of the HEVC video compression algorithm - motion estimation, resulting in lower latency. Development of a machine learning model which utilizes such information to predict motion vectors from one frame in reference to another is explored, as well as its inference in an existing open source HEVC encoder.

• Keywords: bespilotna letjelica; pogled iz prvog lica; HEVC; video kompresija; procjena pokreta; PyTorch; ONNX; ONNX Runtime
• Keywords (english): unmanned aerial vehicle (UAV); first-person view (FPV); HEVC; video compression; motion estimation; PyTorch; ONNX; ONNX Runtime
• Language: croatian
• URN:NBN: urn:nbn:hr:168:388103
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2024-10-12 20:02:37