Poboljšanje guste estimacije stereo dubine korištenjem informacija iz prethodnih slika

master's thesis

• Title (english): Enhancing dense stereo depth estimation via prior image frames information
• Author: Antea Hadviger
• Mentor: Ivan Petrović (mentor)
• Committee member: Ivan Petrović (predsjednik povjerenstva)
• Committee member: Marija Seder (član povjerenstva)
• Committee member: Ivan Marković (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2018-02-20, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Estimacija dubine vrlo je važna za različite primjene u području računalnog vida, uključujući robotiku i autonomnu vožnju. Algoritmi za ostvarivanje stereo korespondencije računaju mape dispariteta uparujući odgovarajuće točke na dvjema slikama dobivenima pomoću stereo kamera. Pokazano je kako algoritam SGM može biti efikasno implementiran koristeći SIMD instrukcije i višedretvenost. Složenost i točnost algoritma može se poboljšati korištenjem informacija iz prethodnih scena. Mapa dispariteta prethodne scene transformira se u prostor trenutne scene koristeći estimaciju gibanja pomoću vizualne odometrije. Nova mapa dispariteta računa se pretražujući smanjeni prostor dispariteta temeljen na predikciji. Obje mape dispariteta na kraju se kombiniraju koristeći Kalmanov filtar. Iako predložena metoda najbolje funkcionira na statičnim scenama, može se proširiti tako da obavlja i detekciju gibajućih objekata. Sve su metode validirane na KITTI podatkovnom skupu.

Abstract (english)

Depth estimation is required for many computer vision applications, including robotics and autonomous driving. Stereo matching algorithms estimate the disparity maps by matching corresponding points in the two images obtained by the stereo cameras. It is presented how the semi-global matching method can be efficiently implemented using SIMD instructions and multithreading. Computational complexity and accuracy of stereo matching can be improved by using information from the previous frames. The disparity map of the previous frame is transformed into the current frame using the ego-motion estimation relying on visual odometry. The newly matched disparity map is calculated using the reduced disparity search space based on the prediction and both disparity maps are combined using the Kalman filter. Although this method performs best on static scenes, it can be extended to detect moving objects as well. The proposed methods are validated on real-world data from the KITTI benchmark suite.

• Keywords: računalni vid; stereo vid; vizualna odometrija; estimacija dubine; detekcija gibajućih objekata
• Keywords (english): computer vision; stereo vision; visual odometry; depth estimation; moving objects detection
• Language: english
• URN:NBN: urn:nbn:hr:168:539424
• 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: 2019-03-13 20:45:46