ETRI-Activity3D: A Large-Scale RGB-D Dataset for Robots to Recognize Daily Activities of the Elderly

Jang, Jongsoo; Kim, Dohyung; Park, Cheonshu; Jang, Minsu; Lee, Jaeyeon; Kim, Jae Hong

doi:10.1109/iros45743.2020.9341160

Cited by 50 publications

(58 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Despite its controlled environment, this large-scale dataset helps to improve deep learning techniques [2,[24][25][26][27][28][29][30][31][32][33] with various approaches. The ETRI-Activity3D dataset [20] is also the multimodal dataset, which provides the actions of the eldery, and KIST SynADL [21] generated synthetic data with motion capture data. It was possible to recognize target actions by the release of the datasets, but there was no guarantee that it would recognize the target actions in different environments with high accuracy.…”

Section: Related Workmentioning

confidence: 99%

“…Due to the need for target action data, datasets containing daily activities such as NTU RGB+D [18,19], ETRI-Activity3D [20], and KIST SynADL [21] have been actively released. They were produced on purpose by directly capturing actions for actual use instead of collecting existing videos.…”

Section: Introductionmentioning

confidence: 99%

“…To handle this problem, we introduce a new dataset of ETRI-Activity3D-LivingLab captured for testing on actual environments. The new dataset has the same class configuration as ETRI-Activity3D [20] and KIST SynADL [21], but has been produced in diverse environments different from the other datasets. Through action recognition models trained on ETRI-Activity3D [20] and KIST SynADL [21] and tested on ETRI-Activity3D-LivingLab, the performance degradation problem due to environmental changes can be observed.…”

Section: Introductionmentioning

confidence: 99%

“…We introduce a new ETRI-Activity3D-LivingLab dataset with tremendous diversity, which does not exist in the ETRI-Activity3D [20] and KIST SynSDL [21] datasets. We reveal the dataset shift problem between datasets and propose a new training and testing protocol to target it.…”

mentioning

confidence: 99%

“…We present various experimental results and analyses on ETRI-Activity3D [20], KIST SynADL [21], and ETRI-Activity3D-LivingLab. The experimental results demonstrate that the close-up of the maximum activation process effectively reduces the dataset shift problem by using various methods.…”

mentioning

confidence: 99%

See 4 more Smart Citations

Action Recognition Using Close-Up of Maximum Activation and ETRI-Activity3D LivingLab Dataset

Kim

Lee

Kim

et al. 2021

Sensors

Self Cite

View full text Add to dashboard Cite

The development of action recognition models has shown great performance on various video datasets. Nevertheless, because there is no rich data on target actions in existing datasets, it is insufficient to perform action recognition applications required by industries. To satisfy this requirement, datasets composed of target actions with high availability have been created, but it is difficult to capture various characteristics in actual environments because video data are generated in a specific environment. In this paper, we introduce a new ETRI-Activity3D-LivingLab dataset, which provides action sequences in actual environments and helps to handle a network generalization issue due to the dataset shift. When the action recognition model is trained on the ETRI-Activity3D and KIST SynADL datasets and evaluated on the ETRI-Activity3D-LivingLab dataset, the performance can be severely degraded because the datasets were captured in different environments domains. To reduce this dataset shift between training and testing datasets, we propose a close-up of maximum activation, which magnifies the most activated part of a video input in detail. In addition, we present various experimental results and analysis that show the dataset shift and demonstrate the effectiveness of the proposed method.

show abstract