LSTM-Based Real-Time Action Detection and Prediction in Human
Motion Streams

J 2019

LSTM-Based Real-Time Action Detection and Prediction in Human Motion Streams

CARRARA, Fabio, Petr ELIÁŠ, Jan SEDMIDUBSKÝ and Pavel ZEZULA

Basic information

Original name

LSTM-Based Real-Time Action Detection and Prediction in Human Motion Streams

Authors

CARRARA, Fabio (380 Italy), Petr ELIÁŠ (203 Czech Republic, belonging to the institution), Jan SEDMIDUBSKÝ (203 Czech Republic, guarantor, belonging to the institution) and Pavel ZEZULA (203 Czech Republic, belonging to the institution)

Edition

Multimedia Tools and Applications, Springer US, 2019, 1380-7501

Other information

Language

English

Type of outcome

Článek v odborném periodiku

Field of Study

10200 1.2 Computer and information sciences

Country of publisher

Netherlands

Confidentiality degree

není předmětem státního či obchodního tajemství

References:

URL

Impact factor

Impact factor: 2.313

RIV identification code

RIV/00216224:14330/19:00109721

Organization unit

Faculty of Informatics

DOI

http://dx.doi.org/10.1007/s11042-019-07827-3

UT WoS

000485298000024

Keywords in English

motion capture data;stream annotation;action detection and recognition;action prediction;LSTM;recurrent neural network

Abstract

V originále

Motion capture data digitally represent human movements by sequences of 3D skeleton configurations. Such spatio-temporal data, often recorded in the stream-based nature, need to be efficiently processed to detect high-interest actions, for example, in human-computer interaction to understand hand gestures in real time. Alternatively, automatically annotated parts of a continuous stream can be persistently stored to become searchable, and thus reusable for future retrieval or pattern mining. In this paper, we focus on multi-label detection of user-specified actions in unsegmented sequences as well as continuous streams. In particular, we utilize the current advances in recurrent neural networks and adopt a unidirectional LSTM model to effectively encode the skeleton frames within the hidden network states. The model learns what subsequences of encoded frames belong to the specified action classes within the training phase. The learned representations of classes are then employed within the annotation phase to infer the probability that an incoming skeleton frame belongs to a given action class. The computed probabilities are finally compared against a learned threshold to automatically determine the beginnings and endings of actions. To further enhance the annotation accuracy, we utilize a bidirectional LSTM model to estimate class probabilities by considering not only the past frames but also the future ones. We extensively evaluate both the models on the three use cases of real-time stream annotation, offline annotation of long sequences, and early action detection and prediction. The experiments demonstrate that our models outperform the state of the art in effectiveness and are at least one order of magnitude more efficient, being able to annotate 10k frames per second.

Links

EF16_019/0000822, research and development project

Name: Centrum excelence pro kyberkriminalitu, kyberbezpečnost a ochranu kritických informačních infrastruktur

Citovat

CARRARA, Fabio, Petr ELIÁŠ, Jan SEDMIDUBSKÝ and Pavel ZEZULA. LSTM-Based Real-Time Action Detection and Prediction in Human Motion Streams. Multimedia Tools and Applications. Springer US, 2019, vol. 78, No 19, p. 27309-27331. ISSN 1380-7501. Available from: https://dx.doi.org/10.1007/s11042-019-07827-3.

@article{1533577,
   author = {Carrara, Fabio and Eliáš, Petr and Sedmidubský, Jan and Zezula, Pavel},
   article_number = {19},
   doi = {http://dx.doi.org/10.1007/s11042-019-07827-3},
   keywords = {motion capture data;stream annotation;action detection and recognition;action prediction;LSTM;recurrent neural network},
   language = {eng},
   issn = {1380-7501},
   journal = {Multimedia Tools and Applications},
   title = {LSTM-Based Real-Time Action Detection and Prediction in Human Motion Streams},
   url = {http://dx.doi.org/10.1007/s11042-019-07827-3},
   volume = {78},
   year = {2019}
}

TY  - JOUR
ID  - 1533577
AU  - Carrara, Fabio - Eliáš, Petr - Sedmidubský, Jan - Zezula, Pavel
PY  - 2019
TI  - LSTM-Based Real-Time Action Detection and Prediction in Human Motion Streams
JF  - Multimedia Tools and Applications
VL  - 78
IS  - 19
SP  - 27309-27331
EP  - 27309-27331
PB  - Springer US
SN  - 13807501
KW  - motion capture data;stream annotation;action detection and recognition;action prediction;LSTM;recurrent neural network
UR  - http://dx.doi.org/10.1007/s11042-019-07827-3
L2  - http://dx.doi.org/10.1007/s11042-019-07827-3
N2  - Motion capture data digitally represent human movements by sequences of 3D skeleton configurations. Such spatio-temporal data, often recorded in the stream-based nature, need to be efficiently processed to detect high-interest actions, for example, in human-computer interaction to understand hand gestures in real time. Alternatively, automatically annotated parts of a continuous stream can be persistently stored to become searchable, and thus reusable for future retrieval or pattern mining. In this paper, we focus on multi-label detection of user-specified actions in unsegmented sequences as well as continuous streams. In particular, we utilize the current advances in recurrent neural networks and adopt a unidirectional LSTM model to effectively encode the skeleton frames within the hidden network states. The model learns what subsequences of encoded frames belong to the specified action classes within the training phase. The learned representations of classes are then employed within the annotation phase to infer the probability that an incoming skeleton frame belongs to a given action class. The computed probabilities are finally compared against a learned threshold to automatically determine the beginnings and endings of actions. To further enhance the annotation accuracy, we utilize a bidirectional LSTM model to estimate class probabilities by considering not only the past frames but also the future ones. We extensively evaluate both the models on the three use cases of real-time stream annotation, offline annotation of long sequences, and early action detection and prediction. The experiments demonstrate that our models outperform the state of the art in effectiveness and are at least one order of magnitude more efficient, being able to annotate 10k frames per second.
ER  -

CARRARA, Fabio, Petr ELIÁŠ, Jan SEDMIDUBSKÝ and Pavel ZEZULA. LSTM-Based Real-Time Action Detection and Prediction in Human Motion Streams. \textit{Multimedia Tools and Applications}. Springer US, 2019, vol.~78, No~19, p.~27309-27331. ISSN~1380-7501. Available from: https://dx.doi.org/10.1007/s11042-019-07827-3.

Detailed Information on Publication Record