MotiFitness AI
An intelligent fitness companion integrating real-time posture analysis using computer vision models, auto-generating macro-balanced meal plans, and adapting workout routines.
On-Device Pose Estimation & Health Integration
Project Overview
Incorrect posture during high-intensity training leads to injury and limits physical gains. MotiFitness AI resolves this by using the mobile device camera to provide real-time, on-device joint angle analysis during workouts.
We designed a lightweight pose-estimation model that runs locally on iOS and Android devices, tracking key joints (knees, hips, shoulders) to evaluate squats, deadlifts, and planks. Biometric inputs are mapped to auto-generate macro-balanced nutrition guides, providing a cohesive training dashboard.
Key Deliverables
- Real-time posture analysis using lightweight MobileNet pose tracking models.
- Bespoke macro-balanced diet generation adapting to biometric sensor updates.
- Integrated health tracking connected to Apple HealthKit and Google Fit APIs.
- Dynamic workout sequence generator adapting to active user fatigue signals.
- Offline-first client sync enabling exercise tracking in zones without connectivity.
The Challenge
Running real-time, frame-by-frame joint classification models on standard mobile processors generally triggers severe thermal throttling, high battery drain, and frame latency delays (under 15fps).
The Solution
Ported MobileNet weights using PyTorch Mobile, writing custom shaders to leverage GPU CoreML (iOS) and NNAPI (Android) architectures. This locked the coordinate parsing pipeline to a stable 30fps at less than 12% CPU capacity.
Measurable Results
Interested in Similar Capabilities?
Discuss mobile computer vision integrations, lightweight ML deployment, or health integrations with our design engineers.