🚦 Traffic Monitoring in Bangladesh

📘 Project Overview

This project is focused on real-time object detection in traffic video feeds from Dhaka, Bangladesh. Using the YOLO (You Only Look Once) object detection model, I detect and classify vehicles and pedestrians from both images and video data. The project walks through each key stage—from video frame extraction to custom model training—providing a hands-on understanding of how to build, train, and deploy object detection systems.

---

📚 Table of Contents

• Lesson 1: Working with File System Paths
• Lesson 2: Working with Image and Video Data
• Lesson 3: Object Detection with YOLO
• Lesson 4: Training YOLO for Custom Classes
• Lesson 5: Data Augmentation for Object Detection
• Lesson 6: The Self-Driving Trolley Problem
• License
• Contact

---

📘 Lesson 1: Working with File System Paths

🮾 Summary

In this lesson, I learned how to work with file system paths using Python's pathlib module. This module provides a modern, object-oriented approach for handling file paths and directory operations.

🎯 Objectives

• Use pathlib to manage file paths
• Create and organize folders
• Traverse directories
• Filter files based on patterns

🧪a Example Code

from pathlib import Path

path = Path("data/videos")
if path.exists():
    print("Path exists!")

for video_file in path.glob("*.mp4"):
    print(video_file.name)

Path("data/frames").mkdir(parents=True, exist_ok=True)

---

📸 Lesson 2: Working with Image and Video Data

🮾 Summary

I explored how to work with image and video data for object detection using a traffic dataset. This involved extracting frames, parsing bounding box annotations, and visualizing object locations.

🎯 Objectives

• Load and organize the dataset
• Extract frames from videos
• Parse XML annotations
• Visualize bounding boxes on images

📚 New Terms

• Bounding Boxes
• Frame Rate
• XML

✅ Conclusion

I can now prepare datasets by extracting and labeling frames from videos, which sets the foundation for object detection.

---

🧠 Lesson 3: Object Detection with YOLO

🮾 Summary

I used the pre-trained YOLO model to perform object detection on both images and videos. This included parsing YOLO outputs and drawing bounding boxes.

🎯 Objectives

• Use YOLO to detect objects in images and videos
• Visualize detection results
• Handle directories and video inputs

📚 New Terms

• Object Detection
• YOLO
• Bounding Boxes
• Normalized Coordinates

✅ Conclusion

I learned how to apply YOLO to real data sources and interpret its output for further use.

---

🧪 Lesson 4: Training YOLO for Custom Classes

🮾 Summary

I fine-tuned a pre-trained YOLO model to detect custom object classes. This involved converting annotations, organizing data, and running training loops.

🎯 Objectives

• Convert XML to YOLO format
• Structure dataset folders
• Handle malformed data
• Train YOLO to detect new classes

📚 New Term

• YAML

✅ Conclusion

I can now train YOLO to detect any object class I care about, using transfer learning and properly structured datasets.

---

🧬 Lesson 5: Data Augmentation for Object Detection

🮾 Summary

I implemented data augmentation techniques to improve model generalization. I used Torchvision to transform images and adjusted bounding boxes accordingly.

🎯 Objectives

• Explore YOLO’s internal augmentation
• Use Torchvision transforms
• Adjust bounding boxes
• Chain multiple augmentation techniques

📚 New Term

• Data Augmentation

✅ Conclusion

Data augmentation has made my dataset more diverse and helped improve training robustness for custom object detection.

---

🚗 Lesson 6: The Self-Driving Trolley Problem

🮾 Summary

I explored the ethical dimensions of AI through the self-driving trolley problem. I reflected on how values encoded in AI can impact real-world outcomes.

🧐 Key Ideas

• AI lacks moral consciousness
• Decisions reflect developer biases
• Global ethical standards are needed

✅ Conclusion

As I build AI systems, I must remain aware of the ethical impact and strive to design models that are fair, transparent, and responsible.

---

📄 License

This project is licensed under the MIT License. See the LICENSE file for details.

---

📬 Contact

Stephen Kinuthia
📧 Email: kinuthiastephen94@gmail.com
🌐 GitHub: github.com/stephenkinuthia-cell

Feel free to reach out for feedback, questions, or collaboration opportunities!