The rigid body gap filling algorithm
Fragment książki (Materiały konferencyjne)
MNiSW
15
WOS
| Status: | |
| Autorzy: | Smołka Jakub, Łukasik Edyta |
| Wersja dokumentu: | Elektroniczna |
| Arkusze wydawnicze: | 0,63 |
| Język: | angielski |
| Strony: | 337 - 343 |
| Web of Science® Times Cited: | 3 |
| Scopus® Cytowania: | 3 |
| Bazy: | Web of Science | Scopus | IEEE Xplore |
| Efekt badań statutowych | NIE |
| Materiał konferencyjny: | TAK |
| Nazwa konferencji: | 9th International Conference on Human System Interactions (HSI) |
| Skrócona nazwa konferencji: | 9th HSI 2016 |
| URL serii konferencji: | LINK |
| Termin konferencji: | 6 lipca 2016 do 8 lipca 2016 |
| Miasto konferencji: | Portsmouth |
| Państwo konferencji: | WIELKA BRYTANIA |
| Publikacja OA: | NIE |
| Abstrakty: | angielski |
| Using an optical passive motion capture system, researchers encounter problems involving marker occlusion. A marker can be covered during motion. This loss of information may cause various problems. For example, the output values of a biomechanical model (such as forces, moments etc.) cannot be computed or are incomplete in recording fragments in which markers are occluded. This paper presents a new, universal gap-filling algorithm for gaps in the trajectories of markers that belong to object segments which may be modelled (by approximation) as rigid bodies. It can fill in gaps in the trajectory of a marker which can be located at the beginning, at the end or in the middle of a recording. In order to assess the algorithm a series of automatic tests were conducted. A set of gapless files with real motion capture data post-processed by an expert was prepared. Then gaps of varying properties were artificially created in the test files. The gaps were filled using the proposed solution. The obtained results were compared to the original files prepared by the expert. The test dataset contains files with three types of movements representing different motion dynamics. The quality of the presented method was assessed for three body segments (head, torso and pelvic) and for three relative generated gap lengths (10%, 20% and 30%). 840 tests with artificially created missing markers were performed. Errors were highest for the torso segment and lowest for the head segment. |
