|Title:||Background separation and mosaic of video sequence|
Hong Kong Polytechnic University -- Dissertations
Department of Electronic and Information Engineering
|Pages:||viii, 70 leaves : ill. ; 30 cm|
|Abstract:||This dissertation explores the separation and mosaic of background in a video sequence with reference to some well developed algorithms. Some well-developed algorithms are modified in this dissertation for improved performances. More precisely, the technique first identifies background and foreground regions based on local motion estimates. Camera motion is then estimated on the background by applying global motion estimation. Finally, after compensating for camera motion, the background content is temporally integrated in long term memory buffer. Block matching with fast pixel decimation searching is used for local motion estimates because of its robustness and resilience to noise as compared with the differential optical flow equation. However, the block-based nature of the technique may result in block artifacts, especially along moving edges. Using smaller block and search window sizes may reduce the artifact effects but it may create another problem if the actual local motion is larger than the search window size. This dissertation will adopt a Gaussian pyramid hierarchical multi-resolution structure for block matching. This multi-resolution scheme in local estimate improves performance in terms computational load and less block artifacts without the need of using large block or search window. A dominant motion is extracted by a clustering of the motion vectors. The pixel blocks moving according to the dominant motion are identified as background, and otherwise as foreground. To minimize block artifacts along moving edges or error in motion estimates due to outlier or noise, this dissertation proposes a robust technique of clustering pixels block as background or foreground based on the surrounding pixel blocks. Camera motion is robustly estimated on the background. There are a variety of well-developed camera motion models in the literature. The commonly used are the simple translational and zoom, the Affine and the Perspective models. More complex model can describe more complex camera motion, but it demands a higher computational complexity and are more sensitive to noise which may result in estimation error. Obviously, the choice of an appropriate model depends on and be adaptive to different camera motion in a video sequence. Therefore instead of fixing a camera model, this dissertation proposes a progressive adaptive technique. First the simple translational component is computed. If the mean absolute error (MAE) in intensity gray level of the current image and the simple translational camera motion-compensated previous image is within a threshold, only simple translational components are used. If the MAE is above the threshold, then more complex model will be used. This method helps to reduce computational effort as well as coding overhead as most of camera motion is of translational only. Once the background has been segmented and the camera motion estimated, the background information is temporally integrated by mosaicking. To minimize the burring out problem due to noise or motion estimation error when more number of frames accumulate, this dissertation proposes a First-In-First-Out approach on the storage of background pixel intensity under a reference coordinate. The mosaic is dynamically updated with the median of pixel intensity in the memory buffer for the same reference pixel coordinate. The using of pixel intensity buffer has the benefit of providing a more robust mosaic against outlier and noise in forming the background pixels.|
|Rights:||All rights reserved|
Files in This Item:
|b15249608.pdf||For All Users (off-campus access for PolyU Staff & Students only)||3.4 MB||Adobe PDF||View/Open|
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item: