RoadSense

RoadSense is a crowdsourced road quality monitoring system using smartphone sensors and deep learning.

Project Structure

This monorepo is organized into four main components:

📱 /mobile

Tech Stack: React Native (Expo), TensorFlow Lite, React Native Maps.

• app/: Expo Router application screens.
• src/services/:
  • sensor.service.ts: Handles Accelerometer/Gyroscope data collection at 50Hz.
  • tflite.service.ts: Bridge to the TCN-BiLSTM .tflite model for inference.
  • background-tasks/: Logic for running data collection in the background.

🌐 /web

Tech Stack: React.js, Vite, Leaflet.js, Tailwind CSS.

• src/components/Map/:
  • MapContainer.jsx: Main Leaflet map instance.
  • HeatmapLayer.jsx: Visualizes pothole density.
• src/pages/: Admin dashboard views.

🧠 /ml-pipeline

Tech Stack: Python, TensorFlow/Keras.

• raw_data/: Sensor datasets (Kaggle/Collected).
• processed_data/: Windowed time-series data.
• models/:
  • final/: Exported .tflite models for the mobile app.
• src/: Training scripts for the TCN-BiLSTM model.

🗄️ /backend

Tech Stack: Supabase (PostgreSQL + PostGIS).

• supabase/migrations/: SQL definitions for geospatial tables.
• supabase/functions/: Edge functions.

Implementation Phases

1. ML Pipeline: Train and export the model.
2. Mobile: specific sensor collection and TFLite integration.
3. Integration: Connect ML model to real-time mobile data.
4. Web: Visualize results.

Contributers

• Contributed by Ayush Kumar