SEDMIDUBSKÝ, Jan and Pavel ZEZULA. Probabilistic Classification of Skeleton Sequences. In S. Hartmann et al. 29th International Conference on Database and Expert Systems Applications (DEXA 2018). Switzerland: Springer. p. 50-65. ISBN 978-3-319-98811-5. doi:10.1007/978-3-319-98812-2_4. 2018.
Other formats:   BibTeX LaTeX RIS
Basic information
Original name Probabilistic Classification of Skeleton Sequences
Authors SEDMIDUBSKÝ, Jan (203 Czech Republic, guarantor, belonging to the institution) and Pavel ZEZULA (203 Czech Republic, belonging to the institution).
Edition Switzerland, 29th International Conference on Database and Expert Systems Applications (DEXA 2018), p. 50-65, 16 pp. 2018.
Publisher Springer
Other information
Original language English
Type of outcome Proceedings paper
Field of Study 10200 1.2 Computer and information sciences
Confidentiality degree is not subject to a state or trade secret
Publication form storage medium (CD, DVD, flash disk)
Impact factor Impact factor: 0.402 in 2005
RIV identification code RIV/00216224:14330/18:00100948
Organization unit Faculty of Informatics
ISBN 978-3-319-98811-5
ISSN 0302-9743
Doi http://dx.doi.org/10.1007/978-3-319-98812-2_4
UT WoS 000460551600004
Keywords in English motion capture data; nearest-neighbor search; action recognition; action classification; re-ranking; similarity measure
Tags DISA, firank_B
Tags International impact, Reviewed
Changed by Changed by: RNDr. Pavel Šmerk, Ph.D., učo 3880. Changed: 30/4/2019 08:16.
Abstract
Automatic classification of 3D skeleton sequences of human motions has applications in many domains, ranging from entertainment to medicine. The classification is a difficult problem as the motions belonging to the same class needn't be well segmented and can be performed by subjects of various body sizes in different styles and speeds. The state-of-the-art recognition approaches commonly solve this problem by training recurrent neural networks to learn the contextual dependency in both spatial and temporal domains. In this paper, we employ a distance-based similarity measure, based on deep convolutional features, to search for the k-nearest motions with respect to a query motion being classified. The retrieved neighbors are analyzed and re-ranked by additional measures that are automatically chosen for individual queries. The combination of deep features, dynamism in the similarity-measure selection, and a new kNN classifier brings the highest classification accuracy on a challenging dataset with 130 classes. Moreover, the proposed approach can promptly react to changing training data without any need for a retraining process.
Links
GBP103/12/G084, research and development projectName: Centrum pro multi-modální interpretaci dat velkého rozsahu
Investor: Czech Science Foundation
PrintDisplayed: 28/3/2024 12:21