Investigation of different skeleton features for CNN-based 3D action recognition

Ding, Zewei; Wang, Pichao; Ogunbona, Philip; Li, Wanqing

doi:10.1109/icmew.2017.8026286

Cited by 29 publications

(11 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Historically, algorithms designed for skeleton based HAR have mostly focused on creating models that capture the geometric attributes derived from the sequential and spatial properties of skeleton sequences. Previous research commonly employed algorithms such as support vector machines [5], hidden Markov models [4], and dynamic temporal distortion [6]. Deep learning algorithms, which can autonomously acquire features from sizable datasets, eventually eclipsed these.…”

Section: Unimodal Harmentioning

confidence: 99%

“…The skeleton [6] or depth sequence and RGB video exhibit distinct properties in the context of HAR. The skeletal sequence includes the time information of human limb movements.…”

Section: Introductionmentioning

confidence: 99%

“…RGB video processing methods employ Convolutional Neural Networks (CNN) or Long Short-Term Memory (LSTM) to analyse both spatial and temporal features [6]. Although RGB video does carry temporal information, it is simpler and more comprehensive than the skeletal modality since it is more vulnerable to environmental factors.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Multi-Modal Human Action Recognition in videos using LSTM with Feature Fusion

Bharathi,

Sri

2024

Preprint

View full text Add to dashboard Cite

Over past few decades, Human action recognition (HAR) in RGB-Depth videos has been extensively studied on the availability of economical depth sensors. Presently, unimodal methods such as skeleton-based and RGB video-based have achieved significant advancements by utilizing progressively larger amounts of data. Nevertheless, there has been a challenge of extracting the temporal feature in RGB video, spatial feature in skeleton sequence. To solve this challenge, we propose Skeleton-RGB Feature Fusion LSTM (SRF-LSTM) network to combine the skeleton and RGB stream modalities and to recognize actions. The fusion of RGB video and skeleton sequences capture both temporal and spatial information effectively. The proposed SRF-LSTM model performs better with an average accuracy of 90.4% than state-of-the-art techniques on NTU RGB-D dataset.

show abstract

Section: Unimodal Harmentioning

confidence: 99%

“…The skeleton [6] or depth sequence and RGB video exhibit distinct properties in the context of HAR. The skeletal sequence includes the time information of human limb movements.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Multi-Modal Human Action Recognition in videos using LSTM with Feature Fusion

Bharathi,

Sri

2024

Preprint

View full text Add to dashboard Cite

show abstract

“…Usually, 3D skeleton sequence data are converted from vector sequences to pseudo-images in order to meet the needs of CNN input. However, it is usually not easy to represent information with both space and time, so many researchers encode bone joints as multiple two-dimensional pseudo-images and then input them into CNN to learn useful features [1,2].…”

Section: Cnnmentioning

confidence: 99%

Graph Neural Networks for Skeleton-based action recognition

Chen,

Yang,

Yang

2024

AETR

View full text Add to dashboard Cite

One of the important directions of the application of artificial intelligence based on human bone behavior recognition is also a research hotspot in the field of computer vision in recent years. Human image video not only contains complex backgrounds, but also uncertain factors such as changes in illumination and changes in the appearance of the human body, which makes behavior recognition based on image videos have certain limitations. Compared with image video, human skeleton video can well overcome the influence of these uncertain factors, so be- havior recognition based on human skeleton has received more and more attention. The human skeleton sequence not only contains the tempo- ral features, but also the spatial structure features of the human body. How to effectively extract the discriminative spatial and temporal fea- tures from the human skeleton sequence is a problem to be solved. In recent years, many methods have been applied to bone-based behavior recognition, such as Recurrent Neural Network (RNN), Convolutional Neural Network (CNN) and Graph Neural Network (GCN). This article will introduce the content and characteristics of these three methods one by one. , And conduct a comparative analysis on it.

show abstract

“…Convolutional Neural Networks have also been frequently used to encode spatio-temporal features of the skeleton sequences. The earlier approaches have transformed skeleton sequences into image representations to exploit CNNs similarly to Computer Vision applications [50,221]. Another effective technique for processing skeleton data proposed in 2018 is the so-called co-occurrence feature learning networks [125] that apply convolutions over different dimensions of data and features to gradually extract and aggregate spatial and temporal signals.…”

Section: Chapter 2 Deep Learning For Human Activity Recognitionmentioning

confidence: 99%

Feature representation learning for human activity recognition

Khaertdinov

View full text Add to dashboard Cite

Investigation of different skeleton features for CNN-based 3D action recognition

Cited by 29 publications

References 13 publications

Multi-Modal Human Action Recognition in videos using LSTM with Feature Fusion

Multi-Modal Human Action Recognition in videos using LSTM with Feature Fusion

Graph Neural Networks for Skeleton-based action recognition

Feature representation learning for human activity recognition

Contact Info

Product

Resources

About