Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/4823F540-622A-46D8-A0EB-F72D9B4E5266 | ||
| Year | 2018 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Filippos Markolefas, "Virtual video synthesis for personalized training", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2018 https://doi.org/10.26233/heallink.tuc.75311 | ||
| Appears in Collections |
Online personal training allow users to work out from the comfort of their own homes, using workout videos from fitness instructors. Moreover, users of such applications can also use their PC camera and work out with others in a group setting, which carries a plethora of intertwined benefits. Towards training efficiency enhancement and consistency, it could be helpful to extract human silhouette from a video which could be used afterwards on exposure to fitness instructor’s video and make the fitness instructions easily followed by the user. In this thesis, initially we involved with traditional background estimation approaches as well as foreground extraction techniques for videos from static cameras. Unfortunately, none of the methods was able to face efficiently all the possible challenges, including slow moving foreground object and presence of foreground object during background initialization, which are dominant problems in our main video (yoga). Thus, we propose a series of techniques that consist of an initial background reconstruction method followed by a selective update scheme. The background image adaptively converges to ground truth data using the above scheme which combines information of detected moving regions (temporal) and color-based regions (spatial) of a video frame. Finally, we apply the proposed method in different environmental conditions such as video from surveillance cameras and we measure the efficiency of the proposed method.
Copyright © DIAS 2013
