Szczegóły publikacji

Opis bibliograficzny

Autorzy (2)

• AGHTreliński Jacek
• AGHKwolek Bogdan

Słowa kluczowe

Dane bibliometryczne

ID BaDAP	134726
Data dodania do BaDAP	2021-07-19
DOI	10.1007/978-3-030-77967-2_9
Rok publikacji	2021
Typ publikacji	materiały konferencyjne (aut.)
Otwarty dostęp
Wydawca	Springer
Konferencja	21st International Conference on Computational Science
Czasopisma/serie	Lecture Notes in Computer Science, Theoretical Computer Science and General Issues

Abstract

In this paper we present an approach for embedding features for action recognition on raw depth maps. Our approach demonstrates high potential when amount of training data is small. A convolutional autoencoder is trained to learn embedded features, encapsulating the content of single depth maps. Afterwards, multichannel 1D CNN features are extracted on multivariate time-series of such embedded features to represent actions on depth map sequences. In the second stream the dynamic time warping is used to extract action features on multivariate streams of statistical features from single depth maps. The output of the third stream are class-specific action features extracted by TimeDistributed and LSTM layers. The action recognition is achieved by voting in an ensemble of one-vs-all weak classifiers. We demonstrate experimentally that the proposed algorithm achieves competitive results on UTD-MHAD dataset and outperforms by a large margin the best algorithms on 3D Human-Object Interaction Set (SYSU 3DHOI).

Publikacje, które mogą Cię zainteresować