Autonom Car Dev

This repository provides a modular, deep learning-based perception system for autonomous vehicles. It includes solutions for lane detection, road segmentation, person and car detection, traffic sign recognition, and traffic light detection. Each module is self-contained and can be used independently or as part of the integrated pipeline.

---

Table of Contents

• Project Overview
• Folder Structure
• Installation
• Usage
• Modules
• Model and Video Files
• Dataset
• Results and Metrics
• Project Report
• Contributing
• License

---

Project Overview

This project aims to provide a comprehensive perception stack for autonomous driving, leveraging state-of-the-art deep learning models for:

• Lane Detection
• Road Segmentation
• Person and Car Detection
• Traffic Sign Recognition
• Traffic Light Detection

The system is designed for both real-time and offline (video-based) analysis. Each module can be trained, evaluated, and used independently, or combined using the main execution script.

---

Folder Structure

Autonom-Car-Dev/
│
├── execute models/
│   ├── main.py
│   ├── lane_segmentation.pt
│   ├── person_and_car_detection.pt
│   ├── road_segmentation.pt
│   ├── traffic_sign.pt
│   ├── traffic_light.pt
│   ├── output_processed.mp4
│   └── utils/
│
├── Lane Det/
│   ├── Lane Segmentation.ipynb
│   ├── train_batches.png
│   ├── F1_Confidence_Curve.png
│   ├── confusion_matrix_normalized.png
│   └── result.png
│
├── Person and Car Detection/
│   ├── detect.py
│   └── coco_classes.txt
│
├── Road Segmentation/
│   ├── Road Segmentation.ipynb
│   ├── train_batches.png
│   ├── PR_Curve.png
│   ├── conf_matrix_norm.png
│   └── results.png
│
├── Traffic Sign/
│   ├── Traffic Sign Detection.ipynb
│   ├── train_batches.png
│   ├── PR_and_F1_Curve.png
│   ├── conf_matrix_norm.png
│   └── results.png
│
├── Traffic Light/
│   ├── Traffic Light Detection.ipynb
│   ├── train_batches.png
│   ├── PR_and_F1_Curve.png
│   ├── conf_matrix_norm.png
│   └── results.png
│
├── Representation.pdf
└── README.md

---

Installation

1. Python

   • Requires Python 3.7 or higher.

2. Dependencies

   • Install required libraries:

     pip install -r requirements.txt

   • If requirements.txt is not available, main dependencies include:
     • torch
     • torchvision
     • opencv-python
     • numpy
     • matplotlib
     • (and others as used in the notebooks)

---

Usage

1. Download Model Files

Model files (.pt) are not included in this repository due to their size. Download them from below and place them in the execute models/ directory:

• lane_segmentation.pt
• person_and_car_detection.pt
• road_segmentation.pt
• traffic_sign.pt
• traffic_light.pt

2. (Optional) Download Output Video

The processed output video (output_processed.mp4) is also not included due to GitHub's file size limits. Download it from [VIDEO_LINK] and place it in the execute models/ directory.

3. Run the Main Pipeline

To run the integrated perception pipeline:

python "execute models/main.py"

4. Run Individual Modules

Each module has its own Jupyter notebook for training, evaluation, and visualization:

• Lane Detection: Lane Det/Lane Segmentation.ipynb
• Road Segmentation: Road Segmentation/Road Segmentation.ipynb
• Traffic Sign Recognition: Traffic Sign/Traffic Sign Detection.ipynb
• Traffic Light Detection: Traffic Light/Traffic Light Detection.ipynb
• Person and Car Detection: Person and Car Detection/detect.py

---

Modules

Lane Detection

• Deep learning-based lane segmentation.
• Training and evaluation in the notebook.

Results:

output of lane segmentation.

Normalized confusion matrix for lane segmentation model.

---

Road Segmentation

• Semantic segmentation of drivable road area.
• Training, evaluation, and metrics in the notebook.

Results:

output of road segmentation.

Normalized confusion matrix for road segmentation model.

Precision-Recall curve for road segmentation.

---

Person and Car Detection

• Object detection using pre-trained models (e.g., YOLO).
• detect.py for inference; coco_classes.txt for class labels.

---

Traffic Sign Recognition

• Detection and classification of traffic signs.
• Notebook includes training, evaluation, and visualization.

Results:

output of traffic sign detection.

Normalized confusion matrix for traffic sign detection.

Precision-Recall and F1 Score curves for traffic sign detection.

---

Traffic Light Detection

• Detection and state classification (red/yellow/green) of traffic lights.
• Notebook includes all steps and result images.

Results:

output of traffic light detection.

Normalized confusion matrix for traffic light detection.

Precision-Recall and F1 Score curves for traffic light detection.

---

Model and Video Files

• Model Files (.pt): Not included in the repository. Download from [https://drive.google.com/drive/folders/1WnN1khMFUZvGs2huXPGwB-RgITNsT6x8?usp=sharing].
• Output Video: Not included due to size. Download from [https://drive.google.com/file/d/1kjLw8CYi9w-BCvs8nGK25XrdhimGQwLF/view?usp=sharing].

Dataset

• Note: The dataset used in this project could not be uploaded to GitHub due to its large size. However, it was previously shared via email.

---

Results and Metrics

Each module directory contains:

• Result Images: Visualizations of predictions and outputs.
• Metric Plots: Curves and matrices (e.g., F1, PR, confusion matrix) for model evaluation.

---

Project Report

• Representation.pdf: Contains a detailed project report, methodology, and results summary. Recommended for an in-depth understanding of the project.

---

Contributing

1. Fork the repository and create a new branch.
2. Make your changes.
3. Submit a pull request.

---

License

MIT License

TEAM MEMBERS

1. Ömer Panay
2. Zeynep Hanife Çelikoğlu
3. Ramazan Ediz