YOLOv8 and SAM Integration for Real-Time Pothole Detection and Segmentation

Description

This project demonstrates how to use a custom-trained YOLOv8 model for detecting potholes and integrate it with the Segment Anything Model (SAM) for precise, zero-shot segmentation of potholes. It also computes and displays the area of each pothole segment in real-time video footage captured from a car dashcam.

Project Workflow:

1. Custom training of YOLOv8 for pothole detection.
2. Real-time pothole detection on dashcam video.
3. Integration of YOLOv8 with Meta's Segment Anything Model (SAM) for segmentation.
4. Real-time area calculation and annotation of pothole areas.

Prerequisites

• Python 3.8+
• YOLOv8 (Ultralytics)
• Segment Anything Model (SAM) from Meta AI

Steps to Run

1. Install Dependencies

   pip install ultralytics segment_anything opencv-python matplotlib  numpy

2. YOLO and SAM Model Preparation

   • Place your custom YOLO model (best.pt) and SAM model checkpoint (sam_b.pt) in the project folder.

3. Running the Application

   • Execute the provided Jupyter Notebook (yolo-sam.ipynb) or Python script to perform inference on video.

License

This project is for educational purposes.

Authors

• Your Name: Dr T.K.Senthil Kumar

---