partially complete
Po-Go (Posture-good) is a machine learning-powered posture correction system designed for real-time ergonomic health monitoring with on-device ML processing, and Wi-Fi/Bluetooth connectivity. It operates without cloud dependency, ensuring faster, secure, and privacy-focused posture analysis.
Key features:
- Intelligent posture detection using ML-based skeletal tracking for real-time corrections
- Edge processing that eliminates cloud dependency for low-latency analysis
- Multi-mode alerts with IoT speaker alerts and notifications
- Seamless connectivity through Wi-Fi for remote tracking and Bluetooth for localized alerts
Po-Go is not just a posture tracker, it is a next-generation digital wellness companion, engineered to prevent postural fatigue, musculoskeletal issues, and workplace discomfort.
This project involves the creation of a smart posture correction system comprising two components:
- Hardware/IoT Device: An AI-powered camera system designed to monitor and analyze sitting posture.
- Companion App: A mobile and desktop application for configuration, notifications, and metrics tracking.
Poor posture is a common issue leading to long-term health problems such as back pain, spinal misalignment, and reduced productivity. Existing solutions often compromise privacy by relying on cloud-based analysis or lack comprehensive features for health improvement.
This solution emphasizes privacy by performing on-device detection. It incorporates advanced posture tracking, health insights, and smart home integrations, creating a holistic system for improving ergonomics and back health.
- Hardware:
- Raspberry Pi 3 or similar for processing.
- RGB and optional thermal cameras for detection.
- Sound module for audio notifications.
- Wi-Fi and Bluetooth modules for connectivity.
- Software:
- MediaPipe or TensorFlow Lite for pose detection.
- Flutter for cross-platform companion app development.
- APIs for optional cloud storage and smart home integration.
Blueprint/Architecture (mermaid)
flowchart TD
A[IoT Device] -->|Capture Posture Data| B[On-Device Analysis]
B -->|Detect Posture| C{Correct Posture?}
C -- No --> D[Wait 10-20 Seconds]
D -->|Still Incorrect?| E[Send Alert]
D -- Correct --> F[Track Metrics]
E --> G[Sound Notification]
E --> H[App Notification]
H --> I[Desktop/Mobile]
F --> J[Store Metrics]
J --> K[User Profile]
K --> L[Health Insights]
L --> M[Ergonomics Suggestions]
M --> K
H --> N[Smart Home Notification]
- Hardware Setup:
- Assemble the Raspberry Pi with the required cameras and sound module.
- Configure Wi-Fi/Bluetooth for connectivity.
- Software Development:
- Implement on-device posture detection using MediaPipe.
- Develop a mechanism to delay notifications for 10-20 seconds to allow self-correction.
- Develop the companion app for configuration, notifications, and health insights.
- Integration:
- Enable smart home connectivity via webhooks and IFTTT.
- Test the system for accuracy and usability.
- Hardware: Affordable and widely available components such as Raspberry Pi and cameras.
- Software: Leveraging existing frameworks like MediaPipe and Flutter simplifies development.
- Real-Time Performance: Use optimized models and lower frame rates to ensure smooth processing.
- Privacy Concerns: Perform all analysis on-device to eliminate the need for cloud processing.
- Delayed Notifications: Implement a configurable delay (10-20 seconds) to allow for self-correction before sending alerts.
- Multiple User Profiles: Develop a robust profile management system to cater to different users.
- Improved Health: Encourage better posture, reducing the risk of long-term back problems.
- Enhanced Productivity: Promote comfort during work or study sessions.
- Awareness: Provide users with detailed insights into their posture habits and trends.
- Household Adoption: A multi-profile system makes it suitable for families.
- Integration with Daily Life: Compatibility with smart home systems ensures seamless usage.
- Ergonomic Improvements: Educates users on proper sitting posture and provides actionable recommendations.
We found two existing hardware based posture correction solutions:
- UprightPose (Website) - A necklace-type wearable that provides vibratory feedback for posture correction.
- ErgoTac (Paper) - An exoskeleton-based wearable that delivers directional vibrations to assist with posture correction.
- Optimal Price and Performance - Exoskeletons are expensive, and necklace wearables lack full-body insights.
- Non-Intrusive Solution - Po-Go is a purely optical system without discomfort from wearables.
- Enhanced Data Insights - Provides deeper analysis than competitors.
- Uses a webcam and real-time AI analysis for long-term posture tracking.
- No need for physical wearables, reducing discomfort.
- uses AI-powered computer vision via a webcam.
- Monitors full-body posture dynamically.
Advantages:
- No discomfort from wearables.
- Exoskeletons are costly and require training.
- Works seamlessly with any existing setup.
- Tracks posture trends over time with weekly/monthly insights.
- Provides customized alerts when bad posture persists.
Advantages:
- offers long-term analytics.
- Portable and compatible with any desk setup.
- No need to buy expensive hardware—works with a standard webcam and browser.
- Highly scalable—suitable for offices, schools, and remote work setups.
Advantages:
- Lower cost than wearable devices ($100–$200).
- More flexible than AI-based ergonomic chairs.
- Scalable across environments without additional cost.
- MediaPipe Pose Estimation (TensorFlow Lite) for real-time human pose estimation.
- RGB Cameras (webcams, IP cameras) for input.
- Thermal Cameras (optional) for temperature-based posture analysis.
- Edge Processing using AI accelerators for real-time performance.
- Capture Input: RGB/thermal cameras detect body key points.
 |
 |
 |
4. *Posture Alerts & Feedback*: - Alerts triggered if bad posture persists for *20+ minutes*. - *Notifications, posture logs, and AI-based ergonomic advice* provided.
Po-Go operates without cloud processing for privacy, real-time performance, and lower latency.
- On-device AI processing using MediaPipe, OpenCV, TensorFlow Lite.
- No personal data is stored or transmitted—only posture analysis results are sent to the server in JSON format.
- Secure string-based data communication ensures privacy.
| Security Feature | Po-Go’s Approach | Benefit |
|---|---|---|
| On-Device AI | Runs locally on device | Prevents cloud-based data leaks |
| Encrypted Communication | SSL/TLS encryption | Prevents MITM attacks |
| Authentication & Access Control | JWT, RBAC, MFA | Ensures secure user access |
| Minimal Data Storage | No PII, anonymized logging | Eliminates data breach risks |
| Secure Firmware Updates | OTA updates with signature verification | Protects against malware |
- Uses *MediaPipe Pose, which infers *33 3D landmarks from RGB video frames.
- Evaluated using Yoga, Dance, and HIIT datasets.
- Landmarks tracked: shoulders, elbows, wrists, hips, knees, ankles.
- Uses a detector-tracker model inspired by BlazePose GHUM 3D.
- Best accuracy with direct camera angles.
- 3D pose RMSE: ~30mm with stereo cameras.
- BLE Connection to Wearable Device
- Uses HC-05/HM-10 BLE modules to send vibration alerts.
- Audio-Visual Alerts
- Speakers & LED indicators signal posture deviations.
| Feature | Po-Go (Posture Good) | Wearables (e.g., Upright GO 2) | Haptic Feedback (DVFI) |
|---|---|---|---|
| Hardware | None (Uses webcam) | Requires a wearable | Multiple sensors needed |
| User Comfort | Non-intrusive | Potentially uncomfortable | Bulky |
| Cost | Low (software-based) | High (device & adhesives) | Expensive |
| Data Analytics | AI-driven insights | Limited real-time feedback | Limited feedback |
| Customization | Adaptive learning | Fixed sensitivity | Standardized feedback |
| Setup | Easy | Complex (device pairing) | Complex (sensor alignment) |
| Environmental Impact | Eco-friendly | Adhesives & device waste | Multiple components & batteries |
- More Affordable & Accessible: Eliminates extra costs by using webcams and AI software.
- No Wearables = More Comfort: No irritation or bulky accessories.
- Advanced AI & Analytics: Offers real-time tracking and insights.
- Eco-Friendly & Low Maintenance: No batteries, replacements, or recalibration required.
- Easy Setup & Universal Compatibility: Works with any laptop, PC, or smartphone with a webcam.
By offering an AI-powered, non-invasive, cost-effective, and scalable posture monitoring solution, Po-Go surpasses traditional wearable and hardware-based alternatives. It ensures long-term health benefits through real-time tracking, adaptive AI alerts, and seamless integration with everyday workspaces.