Smart City Pothole & Road Damage Detection

A real-time object detection system using YOLOv11 to identify and classify road damage from dashcam footage, enabling municipalities to prioritize repairs.

Key Technical Features

• Architected a real-time smart city road damage detection system using the highly optimized YOLOv11 Nano architecture, achieving a highly competitive mAP@0.5 of 54.0% across 4 distinct damage classes (Longitudinal Cracks, Transverse Cracks, Alligator Cracks, Potholes) on the complex RDD2022 dataset.
• Achieved exceptional performance on high-severity damage, including a 64.2% mAP@0.5 for Alligator Cracks.
• Overcame severe hardware limitations by engineering a robust cloud-training pipeline on Kaggle, successfully training for 50 epochs over 7.2 hours without timeout interruptions.
• Optimized model architecture for edge-device deployment, maintaining lightning-fast inference speed (>60 FPS) with a lightweight footprint (5.5MB weights file), enabling cost-effective integration into municipal dashcam fleets.

Setup

1. Install Dependencies

   python -m venv venv
   source venv/bin/activate
   pip install -r requirements.txt

2. Run the Application

   streamlit run app.py

Repository Structure

• src/: Core logic for data, models, and inference.
• notebooks/: Exploratory Data Analysis and Kaggle Evaluation notebooks.
• configs/: YOLO training configuration.
• tests/: Automated tests.
• app.py: Streamlit web application.