Please use this identifier to cite or link to this item:
http://hdl.handle.net/10603/495601
Title: | Vision Based Online Human Tracking Using Dynamic Object Model |
Researcher: | ANSHUL PAREEK |
Guide(s): | VASUDHA ARORA |
Keywords: | Computer Science Computer Science Artificial Intelligence Engineering and Technology |
University: | GD Goenka University |
Completed Date: | 2021 |
Abstract: | This thesis is concerned with an application of visual object tracking where an algorithm newlineapplied in a real-time scenario has to detect and follow a human overcoming all the newlinepractical challenges. This algorithm can be easily applied to any Kinect based robot for newlinetracking human. Challenges encountered in visual tracking appear because of brusque newlineobject motion, pose variation, non-rigid object structure, full/partial occlusion, and newlinesmooth/abrupt camera motion. The thesis overcomes the aforesaid challenges faced during newlinehuman/object tracking along with critical scrutiny of the efficacy of human tracking newlinealgorithms by analyzing assorted visual parameters. newlineThe first step to human tracking is the selection of a feature detector-descriptor algorithm, newlinein this thesis, a complete analysis of the most popular state of art algorithms are done. An newlineexhaustive comparison of AKAZE, BRISK, DAISY, FREAK, KAZE, ORB, SIFT, SURF is newlinedone by applying different transformations and distortions like blurring, noise, intensity newlinevariation, and rotation, to the original image, and experimental results are obtained newlineconcerning it. Matching evaluation parameters like execution time, matching rate, and no. newlineof keypoint features for each algorithm are acquired. Here based on the results obtained, newlineSURF is selected for further work. newlineAfter going through the literature review it is seen that there are inherent shortcomings newlinepresent in the color-based techniques (sensitivity to illumination variations).These are newlineovercome by interest point-based methods that use gradient features such as SIFT or SURF. newlineThe previous work also shows the limitations of interest point-based methods, majorly they newlineare (1) While dealing with all the frames in a video the matching point varies for each newlineframe and finally fade away, and (2) the soaring computational complexity linked with the newlineevaluation of SURF correspondence connecting a couple of images. newline |
Pagination: | |
URI: | http://hdl.handle.net/10603/495601 |
Appears in Departments: | School of Engineering |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
01_title.pdf | Attached File | 13.7 kB | Adobe PDF | View/Open |
02_prelim pages.pdf | 251.73 kB | Adobe PDF | View/Open | |
03_content.pdf | 20.25 kB | Adobe PDF | View/Open | |
04_abstract.pdf | 73.8 kB | Adobe PDF | View/Open | |
05_chapter 1.pdf | 353.89 kB | Adobe PDF | View/Open | |
06_chapter 2.pdf | 1.8 MB | Adobe PDF | View/Open | |
07_chapter 3.pdf | 1.71 MB | Adobe PDF | View/Open | |
08_chapter 4.pdf | 459.05 kB | Adobe PDF | View/Open | |
09_chapter 5.pdf | 510.15 kB | Adobe PDF | View/Open | |
10_annexures.pdf | 158.7 kB | Adobe PDF | View/Open | |
80_recommendation.pdf | 86.97 kB | Adobe PDF | View/Open |
Items in Shodhganga are licensed under Creative Commons Licence Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).
Altmetric Badge: