Back to search

West Virginia University

Selected topics in video coding and computer vision

Abstract

dc:description.abstract

Video applications ranging from multimedia communication to computer vision have been extensively studied in the past decades. However, the emergence of new applications continues to raise questions that are only partially answered by existing techniques. This thesis studies three selected topics related to video: intra prediction in block-based video coding, pedestrian detection and tracking in infrared imagery, and multi-view video alignment.;In the state-of-art video coding standard H.264/AVC, intra prediction is defined on the hierarchical quad-tree based block partitioning structure which fails to exploit the geometric constraint of edges. We propose a geometry-adaptive block partitioning structure and a new intra prediction algorithm named geometry-adaptive intra prediction (GAIP). A new texture prediction algorithm named geometry-adaptive intra displacement prediction (GAIDP) is also developed by extending the original intra displacement prediction (IDP) algorithm with the geometry-adaptive block partitions. Simulations on various test sequences demonstrate that intra coding performance of H.264/AVC can be significantly improved by incorporating the proposed geometry adaptive algorithms.;In recent years, due to the decreasing cost of thermal sensors, pedestrian detection and tracking in infrared imagery has become a topic of interest for night vision and all weather surveillance applications. We propose a novel approach for detecting and tracking pedestrians in infrared imagery based on a layered representation of infrared images. Pedestrians are detected from the foreground layer by a Principle Component Analysis (PCA) based scheme using the appearance cue. To facilitate the task of pedestrian tracking, we formulate the problem of shot segmentation and present a graph matching-based tracking algorithm. Simulations with both OSU Infrared Image Database and WVU Infrared Video Database are reported to demonstrate the accuracy and robustness of our algorithms.;Multi-view video alignment is a process to facilitate the fusion of non-synchronized multi-view video sequences for various applications including automatic video based surveillance and video metrology. In this thesis, we propose an accurate multi-view video alignment algorithm that iteratively aligns two sequences in space and time. To achieve an accurate sub-frame temporal alignment, we generalize the existing phase-correlation algorithm to 3-D case. We also present a novel method to obtain the ground-truth of the temporal alignment by using supplementary audio signals sampled at a much higher rate. The accuracy of our algorithm is verified by simulations using real-world sequences.

Degree

thesis:*
Name thesis:degree_name
PhD
Level thesis:degree_level
Dissertation
Discipline thesis:degree_discipline
Lane Department of Computer Science and Electrical Engineering

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Dai, Congxia

Subjects

dc:subject × 1

Identifiers

dc:identifier.*

Chain of custody

source
Harvested from
West Virginia University
Base URL
researchrepository.wvu.edu/do/oai/
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Dai, Congxia. Selected topics in video coding and computer vision. Dissertation thesis, https://doi.org/10.33915/etd.2591