Date of Award
12-2024
Document Type
Thesis
Degree Name
Master of Science in Engineering (MSE)
Department
School of Computing
Committee Chair/Advisor
Nianyi Li
Committee Member
Huson Smith
Committee Member
Siyu Huang
Abstract
Moving object segmentation in the presence of atmospheric turbulence is a highly challenging task due to the irregular and time-varying distortions induced by the atmospheric turbulence. This thesis presents an unsupervised approach for segmenting moving objects in videos affected by such atmospheric turbulence. The proposed methodology is grounded in a detect-then-grow scheme: the algorithm begins by identifying a small set of moving object pixels (seed points) with high confidence and progressively expanding a foreground mask from these seed points to segment all moving objects. The proposed approach capitalizes on rigid geometric consistency across video frames to disentangle different types of motion, leveraging the Sampson distance to initialize seedling pixels. To ensure the spatio-temporal consistency of the generated masks, the proposed algorithm employs spatial grouping loss and temporal consistency loss during the refinement phase. Unlike traditional methods, the proposed approach is fully unsupervised and does not require any labeled training data, making it highly adaptable to real-world applications where labeled data may be scarce.
To evaluate the effectiveness of the proposed method, this thesis introduces and releases the Dynamic Object Segmentation in Turbulence (DOST) dataset, the first real-captured long-range turbulent video dataset with ground-truth moving object segmentation masks. The proposed algorithm demonstrates strong accuracy and robustness across varying turbulence strengths. Moreover, it can effectively handle multiple moving objects in dynamic scenes. Comparative results on DOST show that the proposed approach outperforms existing state-of-the-art methods in terms of accuracy and robustness under both normal and severe turbulence conditions. This demonstrates the potential of our framework for long-range video analysis in applications like surveillance, environmental monitoring, and remote sensing.
Recommended Citation
Qin, Dehao, "Unsupervised Moving Object Segmentation with Atmospheric Turbulence" (2024). All Theses. 4425.
https://open.clemson.edu/all_theses/4425
Author ORCID Identifier
0000-0002-3463-5405
