YOLO-Car-Detection

Overview

This project implements an autonomous driving application that detects cars in images using the YOLO (You Only Look Once) object detection model. The model processes images and predicts bounding boxes around detected vehicles, making it useful for real-time object detection in autonomous navigation systems.

Features

• Utilizes YOLOv2 for real-time object detection.
• Processes images to identify and localize vehicles.
• Implements Non-Maximum Suppression (NMS) to filter overlapping bounding boxes.
• Uses TensorFlow and Keras for deep learning inference.
• Includes visualization utilities for displaying predictions.

Project Structure

• Autonomous_driving_application_Car_detection.ipynb: The main Jupyter Notebook containing step-by-step implementation of YOLO-based car detection.
• utils.py: Utility functions for:
  • Preprocessing images for YOLO.
  • Filtering predictions based on confidence scores.
  • Applying Non-Maximum Suppression (NMS) to remove redundant bounding boxes.
  • Drawing bounding boxes on detected objects.
  • Evaluating the YOLO model output and making predictions.

Dependencies

To run this project, install the following packages:

pip install tensorflow numpy pandas scipy pillow matplotlib

How to Run

1. Clone this repository and navigate to the project directory.
2. In the repository clone the repository https://github.com/allanzelener/YAD2K/tree/master and move the files and subfolders to the main project directory.
3. Open Autonomous_driving_application_Car_detection.ipynb in Jupyter Notebook.
4. Run all cells to:
   • Load the YOLO model.
   • Preprocess images for detection.
   • Predict car locations.
   • Display the output with bounding boxes.

YOLO Model Details

• Input: Image of size (608, 608, 3) normalized for YOLO processing.
• Output: Predicted bounding boxes with class probabilities.
• Filtering:
  • Boxes with confidence scores below threshold=0.6 are discarded.
  • Intersection over Union (IoU) used to filter overlapping boxes (iou_threshold=0.5).
  • Maximum of 10 bounding boxes per image.

Results

The model successfully detects vehicles in images, outputs bounding boxes, and applies Non-Maximum Suppression to refine predictions. The detected objects are drawn on the images for visualization.

Future Improvements

• Train on additional datasets for improved accuracy.
• Implement real-time video detection.
• Optimize model for deployment on edge devices (e.g., Raspberry Pi, Jetson Nano).

Credits

This project is inspired by deep learning applications in autonomous driving and object detection. The implementation is based on YOLO principles.

License

This project is open-source and available under the MIT License.

---

Author: Chirag N Vijay