Video Analysis for Content-Based Applications
Video Analysis for Content-Based Applications
The analysis of video for semantics presents significant advantages for many content-based applications. An indicator of this is the recent trend towards the extraction of video semantics in the video processing community. Techniques for the extraction of semantics are varied and often application specific. Efficient interaction and interpretation of video is only as good as the underlying video content representation.
This thesis is concerned with the development of video analysis techniques which support the extraction of semantics such as object and event identify from video. A new methodology for video analysis in which video is processed for high-level content through a hierarchy of visual analysis modules is proposed. Each module is constrained by a context which improves the performance of visual recognition. The design, implementation and testing of the new methodology resulted in a video analysis platform called The VCR System (The Video Content Recognition System).
This thesis proposes a novel way to generate and handle video-objects, which is a step towards overcoming some of the problems in current machine vision. Problems in the manipulation of temporal data such as moving objects in video are examined. The problems include temporal correspondence of objects from one frame to the next, object occlusion, and false-positive object detections. The VCR System contains algorithms which will link objects in a generic way, suppress false detection of objects, and predict missing objects (either due to occlusion or missed detections).
The contributions to the area of video analysis are demonstrated and evaluated by developing the VCR System to locate faces in video. A further more substantial demonstration is presented by analysing videos of snooker.
University of Southampton
Chan, Stephen Chi Yee
1becb734-ef4d-4531-a80b-dc7059fabcf9
2001
Chan, Stephen Chi Yee
1becb734-ef4d-4531-a80b-dc7059fabcf9
Chan, Stephen Chi Yee
(2001)
Video Analysis for Content-Based Applications.
University of Southampton, Doctoral Thesis.
Record type:
Thesis
(Doctoral)
Abstract
The analysis of video for semantics presents significant advantages for many content-based applications. An indicator of this is the recent trend towards the extraction of video semantics in the video processing community. Techniques for the extraction of semantics are varied and often application specific. Efficient interaction and interpretation of video is only as good as the underlying video content representation.
This thesis is concerned with the development of video analysis techniques which support the extraction of semantics such as object and event identify from video. A new methodology for video analysis in which video is processed for high-level content through a hierarchy of visual analysis modules is proposed. Each module is constrained by a context which improves the performance of visual recognition. The design, implementation and testing of the new methodology resulted in a video analysis platform called The VCR System (The Video Content Recognition System).
This thesis proposes a novel way to generate and handle video-objects, which is a step towards overcoming some of the problems in current machine vision. Problems in the manipulation of temporal data such as moving objects in video are examined. The problems include temporal correspondence of objects from one frame to the next, object occlusion, and false-positive object detections. The VCR System contains algorithms which will link objects in a generic way, suppress false detection of objects, and predict missing objects (either due to occlusion or missed detections).
The contributions to the area of video analysis are demonstrated and evaluated by developing the VCR System to locate faces in video. A further more substantial demonstration is presented by analysing videos of snooker.
Text
835272.pdf
- Version of Record
More information
Published date: 2001
Identifiers
Local EPrints ID: 464606
URI: http://eprints.soton.ac.uk/id/eprint/464606
PURE UUID: 02e60617-3f59-4b7e-b315-a396cb012986
Catalogue record
Date deposited: 04 Jul 2022 23:50
Last modified: 16 Mar 2024 19:38
Export record
Contributors
Author:
Stephen Chi Yee Chan
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics