IE Master of Business Analytics and Big Data - Fitizens.
FITIZENS aims to enhance the fitness experience for users globally, making exercise more accessible, efficient, and engaging. In this light, we were tasked with developing a model to detect a set of exercises using MLKit and Computer Vision. The architecture and implementation of the FITIZENS project are characterized by a robust, modular design that facilitates real-time exercise detection and analytics. The system architecture combines video input processing, pose estimation, action recognition, and repetition counting into a cohesive framework.
The methodology adopted for this project involved a comprehensive approach, emphasizing the use of pre-trained models and transfer learning. This strategy facilitated rapid development and deployment by capitalizing on existing datasets and algorithms optimized for similar tasks. Specifically, the project leveraged the Kinetics-400 dataset to train models capable of recognizing a wide range of exercises. Furthermore, the methodology incorporated the use of MediaPipe for pose detection and OpenVINO for action recognition, enabling the system to analyze and interpret complex human movements in real-time accurately.
Deliverables: Final Report, Presentation Slides, Code, Live Demo
Corporate Project developed by Mario Bevilacqua, Charles Beyrard, Aldemar Pinzón, and Marcos Ray