ConvST-LSTM-Net: convolutional spatiotemporal LSTM networks for skeleton-based human action recognition

Human action recognition (HAR) emphases on perceiving and identifying the action behavior done by humans within an image/video. The HAR activities include motion patterns and normal or abnormal activities like standing, walking, sitting, running, playing, falling, fighting, etc. Recently, it sparks...

Full description

Saved in:
Bibliographic Details
Published in:International journal of multimedia information retrieval 2023-12, Vol.12 (2), p.34, Article 34
Main Authors: Sharma, Abhilasha, Singh, Roshni
Format: Article
Language:English
Subjects:
Computer Science
Data Mining and Knowledge Discovery
Database Management
Datasets
Fall detection
Human activity recognition
Human body
Image Processing and Computer Vision
Information Storage and Retrieval
Information Systems Applications (incl.Internet)
Multimedia Information Systems
Neural networks
Regular Paper
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Human action recognition (HAR) emphases on perceiving and identifying the action behavior done by humans within an image/video. The HAR activities include motion patterns and normal or abnormal activities like standing, walking, sitting, running, playing, falling, fighting, etc. Recently, it sparks the attention of researchers especially in 3D skeleton sequence. The actions of human can be represented via sequence of motions of skeletal keyjoints, although not all the skeleton keyjoints are informative in nature. Various approaches for HAR are used like LSTM, ConvLSTM, Conv-GRU, ST-LSTM, etc. Thus far, ST-LSTM approaches have shown tremendous performance in 3D skeleton sequence tasks but the detection of irrelevant keyjoints produce noise that deteriorates the performance of the model. So, the intent is to bring attention toward improving the efficacy of the model by focusing on informative keyjoint coordinates only. Therefore, the research paper introduces a new class of spatiotemporal LSTM approaches named as ConvST-LSTM-Net (convolutional spatiotemporal long short-term memory network) for skeleton-based action recognition. The prime focus of proposed model is to identify the informative keyjoints in each frame. The result of extensive experimental analysis exhibits that ConvST-LSTM-Net outperforms the state-of-the-art models on various benchmarks dataset, viz. NTU RGB + D 60, UT-Kinetics, UP-Fall Detection, UCF101, and HDMB51 for skeleton sequence data.
ISSN:2192-6611
2192-662X
DOI:10.1007/s13735-023-00301-9