Welcome to the InfoSci Platform
Virtual Personal Trainer: Fitness Video Recognition Using Convolution Neural Network and Bidirectional LSTM

Virtual Personal Trainer: Fitness Video Recognition Using Convolution Neural Network and Bidirectional LSTM

Anuja Arora, Anu Taneja, Mayank Gupta, Prakhar Mittal
Copyright: © 2021 |Volume: 12 |Issue: 4 |Pages: 21
ISSN: 1947-8208|EISSN: 1947-8216|EISBN13: 9781799861850|DOI: 10.4018/IJKSS.291976
Cite Article Cite Article

MLA

Arora, Anuja, et al. "Virtual Personal Trainer: Fitness Video Recognition Using Convolution Neural Network and Bidirectional LSTM." IJKSS vol.12, no.4 2021: pp.71-91. http://doi.org/10.4018/IJKSS.291976

APA

Arora, A., Taneja, A., Gupta, M., & Mittal, P. (2021). Virtual Personal Trainer: Fitness Video Recognition Using Convolution Neural Network and Bidirectional LSTM. International Journal of Knowledge and Systems Science (IJKSS), 12(4), 71-91. http://doi.org/10.4018/IJKSS.291976

Chicago

Arora, Anuja, et al. "Virtual Personal Trainer: Fitness Video Recognition Using Convolution Neural Network and Bidirectional LSTM," International Journal of Knowledge and Systems Science (IJKSS) 12, no.4: 71-91. http://doi.org/10.4018/IJKSS.291976

Export Reference

Mendeley
Favorite Full-Issue Download

Abstract

The increased interest of users towards healthier lifestyles has motivated the development of a virtual personal trainer application using Android as platform. Despite the availability of numerous fitness apps and gyms, everyone needs proper training at their ease and wishes to monitor calories burnt. Thus, this paper proposes a novel idea of virtual personal trainer applications that recognizes user actions through videos. The video data is processed using convolutional neural network and bidirectional long short-term memory network. The motive of work is to recognize exercise accurately from video and calculate the number of calories expended. The proposed application provides not only detailed information about exercise but also ascertains the correct way of performing exercises as this is a major challenge that users face due to lack of knowledge. The idea is implemented on UCF-101 Action Recognition dataset, and experimental results show significant improvements as compared to baseline methods. This study would benefit users who are fitness enthusiasts and are more prone to gadgets.

Request Access

You do not own this content. Please login to recommend this title to your institution's librarian or purchase it from the IGI Global bookstore.