- Hitesh Kotte
- Nghia Duong-Trung
- Milos Kravcik
In today's world, where developing psychomotor skills poses a significant challenge, cutting-edge technology steps in as an invaluable ally. Our research introduces an innovative approach that harnesses the power of computer vision to track and enhance performance in fitness training. We present a groundbreaking system designed to monitor exercise execution and offer instantaneous feedback on posture during workouts. This real-time guidance facilitates self-correction and boosts motivation, especially in the absence of professional supervision.
Our framework is built upon advanced computer vision technology, employing the highly efficient YOLOv7-pose model to pinpoint human keypoints. This is coupled with a sophisticated tracking system focused on human body topology, providing comprehensive data for immediate posture correction. One of the standout features of our system is its use of transfer learning, which significantly reduces the need for extensive retraining of the model.
Crucially, our application, FITSIGHT, takes this innovation a step further. It enables users to employ a standard webcam as a virtual fitness trainer. This feature is particularly valuable in situations where a physical trainer is unavailable due to various constraints. FITSIGHT stands as a testament to the potential of technology in bridging the gap in physical training, providing users with a reliable and sophisticated tool for self-guided fitness journeys.
This project has been featured in the following publications:
-
FitSight: Tracking and Feedback Engine for Personalized Fitness Training
- Authors: Hitesh Kotte, Florian Daiber, Miloš Kravčík and Nghia Duong-Trung
- Conference: UMAP 2024: In Proceedings of the 32nd ACM Conference on User Modeling, Adaptation and Personalization, July 1–4, 2024, Cagliari, Italy.
- Link to Publication
-
IMPECT-POSE: A Complete Front-end and Back-end Architecture for Pose Tracking and Feedback
- Authors: Abhishek Samantha, Hitesh Kotte, Patrick Handwerk, Khaleel Asyraaf Mat Sanusi, Mai Geisen, Miloš Kravčík, and Nghia Duong-Trung
- Conference: In Proceedings of the 32nd ACM Conference on User Modeling, Adaptation and Personalization, July 1–4, 2024, Cagliari, Italy.
- Link to Publication
-
Augmented Intelligence in Tutoring Systems: A Case Study in Real-Time Pose Tracking to Enhance the Self-learning of Fitness Exercises
- Authors: Nghia Duong-Trung, Hitesh Kotte and Miloš Kravčík
- Conference: 18th European Conference on Technology Enhanced Learning, EC-TEL 2023 Aveiro, Portugal, September 4–8, 2023 Proceedings
- Link to Publication
-
Real-Time Posture Correction in Gym Exercises: A Computer Vision-Based Approach for Performance Analysis, Error Classification and Feedback
- Authors: Hitesh Kotte, Miloš Kravčík and Nghia Duong-Trung
- Conference: MILeS’23: 18th European Conference on Technology Enhanced Learning (ECTEL2023), 3rd International Workshop on Multimodal Immersive Learning Systems, September 04–08, 2023, Aveiro, Portugal.
- Link to Publication
This project was also submitted as a Master thesis:
- Augmented Intelligence in Tutoring Systems: Real-Time Pose Tracking to Enhance the Self-learning of Fitness Exercises
- Authors: Hitesh Kotte
- Supervisor: Dr. Florian Daiber
- Reviewers: Prof. Dr. Antonio Krüger and Dr. Nghia Duong-Trung
- Submitted To: Data Science and Artificial Intelligence department, Saarland University, Germany.
Real Time Feedback |
|
Fitsight Workflow |
Impect Dashboard |
Impect Platform Architecture |
-
Ensure you have Anaconda installed to manage your environments and packages.
-
Clone the project repository from GitHub. Run the following command in your terminal:
git clone https://github.com/hiteshkotte/DFKI-fitsight.git
After cloning, extract the project files to your desired location. Ensure that you have sufficient permissions in the directory where you wish to extract the files.
-
Create a Virtual Environment:
- Open Anaconda Prompt.
- Run the following command to create a new virtual environment named
fitsight
:conda create -n fitsight
-
Activate the Virtual Environment:
- Activate the
fitsight
environment by executing:conda activate fitsight
- Activate the
-
Install Required Packages:
- Ensure that the
fitsight
environment is activated. - Change your current directory to the
DFKI-fitsight
folder, which is your working directory. Run the following command:cd path/to/your/downloads/DFKI-fitsight
Replace
path/to/your/downloads/DFKI-fitsight
with the actual path where theDFKI-fitsight
folder is located on your machine.- Install the required packages by running:
pip install -r requirements.txt
- Ensure that the
-
FFmpeg Installation: FFmpeg is required for handling multimedia files.
-
Using Anaconda Powershell:
- Run the following command to install FFmpeg:
conda install -c conda-forge ffmpeg
- Run the following command to install FFmpeg:
-
For Non-Anaconda Environments:
-
Download and extract the FFmpeg build to the working directory.
-
Add the path of the extracted FFmpeg
bin
folder to your system's PATH:- Locate the
bin
folder inside the extracted FFmpeg directory. - Copy the full path to this
bin
folder. - Add this path to your system's PATH:
- Locate the
-
Add FFmpeg to your system's PATH:
- For all users:
setx /M PATH "%PATH%;your_path"
- For the current user:
setx PATH "%PATH%;your_path"
- For all users:
-
In this version, "your_path" should be replaced by the user with the actual path to the FFmpeg
bin
directory. The command updates the PATH environment variable, making FFmpeg accessible from anywhere in the command line interface.
-
-
-
Run the Application:
- Start the application by running the following command in your terminal or command prompt:
python app.py
- Once the application is running, open your web browser.
- Navigate to the local host address provided in the application's output. Typically, this will be
http://localhost:PORT
- Start the application by running the following command in your terminal or command prompt:
-
Accessing the Application in the Browser:
- When you open the application in the browser, it will prompt you for credentials.
- Use the following 🗝️Login Credentials:
- Username: admin
- Password: admin
This will grant you access to the application's features.
To access the input videos used for exercise analysis, please visit the following Google Drive link:
These videos are essential for testing and demonstrating the functionalities of the application. Ensure you have appropriate permissions to access this content.