Date of Graduation

2016

Document Type

Thesis

Degree Type

MS

College

Statler College of Engineering and Mineral Resources

Department

Lane Department of Computer Science and Electrical Engineering

Committee Chair

Xin Li

Committee Co-Chair

Vinod K Kulathumani

Committee Member

Matthew C Valenti

Abstract

Amateur videos captured by consumer-level devices (e.g., mobile phones, tablets, camcorders etc.) are often shaky, undirected and difficult to watch. Video stabilization techniques attempt to improve the video quality by filtering out unwanted jittering camera motions. Currently there are two key challenges in the field of video stabilization: robustness (e.g., in the presence of sophisticated camera motion including panning and rotation) and quality evaluation. Therefore the main contributions of this thesis are two-fold.;First, we present a robust video stabilization algorithm, which post-processes shaky video data and addresses the challenges of rapid camera panning and motion blur. Our algorithm is based on a robust 2-D motion estimation method using Random Sample Consensus (RANSAC) and M-estimator Sample Consensus (MSAC). Additionally, K-Nearest Neighbor based feature replacement was developed to further improve the robustness of feature tracking. It is experimentally demonstrated that the proposed video stabilization algorithm can handle video with low feature count and motion blur.;Second, we present a Total--Variation (TV) based quality evaluation metric, which objectively quantifies the shakiness in amateur video. The proposed stabilization technique is compared against existing online video stabilization software (e.g., Deshaker and YouTube) using the newly-developed quality metric. Experimental results demonstrate that the developed algorithm is performing both subjectively and objectively at least as well as benchmark methods.

Share

COinS