How dhiwinsamrich/hand_sign_recoginition_iot works
Project Type and Purpose
• This is a full-stack IoT + ML system for real-time hand-sign recognition using a smart glove.
• Key domains: Embedded hardware (sensors + microcontrollers), edge networking, cloud-hosted REST API, and a machine-learning inference service.
File Structure
• Top-level hardware/software separation is implicit:
• “Model/” holds serialized Keras models (.h5) used by the server.
• “server.py” is the Flask REST API that loads the model and exposes a /predict endpoint.
• “IPYNB File/” and standalone .ipynb notebooks encapsulate data exploration, training, and deployment experiments (e.g., “deploy a saved keras model to aws.ipynb”).
• No explicit “frontend” folder— the glove + microcontroller acts as the edge client.
• Deployment artifacts: LICENSE, README.md, certificates, and notebooks for AWS EC2 deployment.
README for Additional Insights
• Hardware: Flex sensors ×5 + MPU-6500 → Arduino Nano → NodeMCU ESP8266 (Wi-Fi).
• ML: A supervised classifier (Random Forest/SVM or a Keras model) with 95.7% accuracy, serialized and hosted in Flask.
• Cloud: AWS EC2 running Flask + Gunicorn, uptime 99.9%.
• Data flow: glove → JSON POST → /predict → JSON response.
Main Components & Interactions
a. Components
1. Hardware Layer
• Flex Sensors + IMU (MPU-6500)
• Arduino Nano (sensor aggregation)
• NodeMCU ESP8266 (Wi-Fi client)
2. Network Layer
• Wi-Fi/Internet (HTTP(S))
3. Cloud/API Layer
• AWS EC2 (Ubuntu)
• Gunicorn + Flask (server.py)
4. ML Inference Layer
• Keras model files (Model/mini_model.h5, my_model.h5)
• In-process inference (via Flask)
5. Clients / Testing
• Postman or any HTTP client
6. (Optional) Offline Training Environment
• Jupyter Notebooks (IPYNB File/) for training, evaluation, and model export
b. Relationships & Data Flow
• Sensor data (flex + gyro + accel) → Arduino Nano serial → NodeMCU packages JSON.
• NodeMCU → HTTP POST to AWS EC2 /predict.
• Flask loads model, runs inference at <150 ms, returns JSON gesture.
• NodeMCU (or end-user) receives the recognized gesture.
• Offline: Data scientist uses notebooks to retrain and update Model/.
c. Architectural Patterns & Principles
• Edge-to-Cloud pattern: perform minimal preprocessing on device, heavy inference in cloud.
• Microservice-style API: single /predict endpoint, stateless inference service.
• Separation of concerns: Hardware firmware vs. network client vs. cloud service vs. ML training.
d. Relevant Technologies & Frameworks
• Firmware: Arduino IDE (C/C++), Serial Monitor
• Edge Client: NodeMCU ESP8266 (MicroPython/Arduino)
• ML: Python, NumPy, Pandas, scikit-learn or Keras (h5 serialization)
• API: Flask + Gunicorn
• Deployment: AWS EC2 (Ubuntu), JSON over HTTP