A Deep Learning Approach for Human Action Recognition Using Skeletal Information

Mathe, Eirini; Maniatis, Apostolos; Spyrou, Evaggelos; Mylonas, Phivos

doi:10.1007/978-3-030-32622-7_9

Cited by 8 publications

(2 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They reported that previously learned CNN-based representation on largescale annotated datasets is effectively transportable to HAR tasks with constrained training datasets. A CNN-based HAR method for ADLs was described by Mathe et al 11 DFT images are used to train a neural network, meaning that all HARs are ultimately represented by images. They generated 3D skeleton locations of human joints from raw RGB sequences and improved them with depth data.…”

Section: Review Of Existing Har Approachesmentioning

confidence: 99%

(Retracted) Modeling of human action recognition using hyperparameter tuned deep learning model

Jain

Neelakandan

et al. 2022

J. Electron. Imag.

View full text Add to dashboard Cite

.Recently, human action recognition (HAR) has become an important focus of computer science research because of its applications in surveillance, robotics, sentiment analysis, and other areas. Human activity classification is a time-consuming operation, especially when photos are cluttered, and the background is unclear. In addition, conventional machine learning models fail to achieve robust performance due to the increasing number of activities. Comparatively, deep learning (DL) approaches help to automatically learn and describe the necessary features of the input data with low manual work and robust discriminant abilities. In addition, a preprocessing stage is included in this study’s HAR utilizing hyperparameter tuned DL (HAR-HPTDL) model that removes undesired background and improves the quality of the input. The model also implements a bidirectional long short-term memory model as a feature extractor, the sparrow search algorithm to tune the hyperparameters, and a SoftMax layer for the effective classification of human actions. In addition, the curse of dimensionality can be overcome via entropy-based feature reduction and Chi square-based feature selection. Based on a variety of measures, the HAR-HPTDL methodology has been put through its paces with other published techniques. The results indicate that the HAR-HPTDL technique outperforms current state-of-the-art techniques in simulations. The outcome of this work demonstrates that an HAR-HPTDL model may achieve comparable or even superior recognition accuracy of 0.949 than the prior best deep classifier(s) on all databases with proper parameter optimization.

show abstract

Section: Review Of Existing Har Approachesmentioning

confidence: 99%

(Retracted) Modeling of human action recognition using hyperparameter tuned deep learning model

Jain

Neelakandan

et al. 2022

J. Electron. Imag.

View full text Add to dashboard Cite

show abstract

“…In this context, deep learning solutions are trained on the data collected from a sensor, or a set of sensors, in order to automatically identify the user's activities [14]. The most attractive deep learning architectures for skeleton-based HAR are Recurrent Neural Networks (RNN) [4,[15][16][17][18][19][20][21][22].…”

Section: Introductionmentioning

confidence: 99%

Recurrent Network Solutions for Human Posture Recognition Based on Kinect Skeletal Data

Guerra

Ramat

Beltrami

et al. 2023

Sensors

View full text Add to dashboard Cite

Ambient Assisted Living (AAL) systems are designed to provide unobtrusive and user-friendly support in daily life and can be used for monitoring frail people based on various types of sensors, including wearables and cameras. Although cameras can be perceived as intrusive in terms of privacy, low-cost RGB-D devices (i.e., Kinect V2) that extract skeletal data can partially overcome these limits. In addition, deep learning-based algorithms, such as Recurrent Neural Networks (RNNs), can be trained on skeletal tracking data to automatically identify different human postures in the AAL domain. In this study, we investigate the performance of two RNN models (2BLSTM and 3BGRU) in identifying daily living postures and potentially dangerous situations in a home monitoring system, based on 3D skeletal data acquired with Kinect V2. We tested the RNN models with two different feature sets: one consisting of eight human-crafted kinematic features selected by a genetic algorithm, and another consisting of 52 ego-centric 3D coordinates of each considered skeleton joint, plus the subject’s distance from the Kinect V2. To improve the generalization ability of the 3BGRU model, we also applied a data augmentation method to balance the training dataset. With this last solution we reached an accuracy of 88%, the best we achieved so far.

show abstract

Artificially Intelligent and Sustainable Smart Cities

Gourisaria

Jee

Harshvardhan

et al. 2022

Studies in Computational Intelligence

View full text Add to dashboard Cite

A Deep Learning Approach for Human Action Recognition Using Skeletal Information

Cited by 8 publications

References 10 publications

(Retracted) Modeling of human action recognition using hyperparameter tuned deep learning model

(Retracted) Modeling of human action recognition using hyperparameter tuned deep learning model

Recurrent Network Solutions for Human Posture Recognition Based on Kinect Skeletal Data

Artificially Intelligent and Sustainable Smart Cities

Contact Info

Product

Resources

About