Publication: Extracting Gait Figures in a Video Based on Markerless Motion
6
item.page.dc.date.issued
2016-01-04
item.page.dc.type
item.page.dc.identifier.doi
item.page.dc.identifier.other
2-s2.0-84964765571
item.page.dc.rights
Mahidol University
item.page.dc.rights.holder
SCOPUS
item.page.dc.identifier.citation
Proceedings - 2015 IEEE International Conference on Knowledge and Systems Engineering, KSE 2015. (2016), 306-309
punsarn.item.page.citation
Worapan Kusakunniran Extracting Gait Figures in a Video Based on Markerless Motion. Proceedings - 2015 IEEE International Conference on Knowledge and Systems Engineering, KSE 2015. (2016), 306-309. doi:10.1109/KSE.2015.16 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/43467
Research Projects
Organizational Units
Authors
Journal Issue
relationships.isThesisOf
item.page.dc.title
Extracting Gait Figures in a Video Based on Markerless Motion
item.page.dc.contributor.author
item.page.dc.contributor.other
item.page.dc.description.abstract
© 2015 IEEE. This paper proposes a new method to extract gait figures in a 2D video without using any markers. Such scenario is more feasible in a real-world environment than a traditional 3D cooperative multicamera system with reflective markers which is costly and complicated. The proposed method is developed to extract following information from a 2D gait video based on marker less motion: 1) a gait period, 2) key positions of a human body (i.e. Head, waist, left-knee, right-knee, left-ankle, and right-ankle) in each frame within a gait period. This is processed by using statistical techniques including linear regression, parabolic regression and polynomial interpolation. Such extracted gait information is useful for many gait-based applications such as human identification in a surveillance system, injury analysis in a sport science, and disease detection and gait rehabilitation in a clinical area. The widely adopted CASIA gait database B is used to verify the proposed method. The extracted key positions are validated by comparing with a ground-truth which is manually generated by human observers. The experimental results demonstrate that the proposed method can achieve very promising performance.