🛣️ Object Detection on Streets Using YOLO and CNNs

This project implements a machine learning system to identify and classify various objects typically found on streets—such as cars, trucks, bicycles, and pedestrians. It leverages both classical and deep learning models, including Convolutional Neural Networks (CNNs) and the YOLO (You Only Look Once) object detection framework.

---

🚀 Features

• Detects objects in both images and videos
• Supports over 80 object classes from the COCO dataset
• Draws bounding boxes with class labels and confidence scores
• Runs on GPU using TensorFlow and Keras for fast inference

---

🧠 Models Used

• YOLO: For fast, accurate, real-time object detection
• CNNs: For low-level image feature extraction and classification
• Linear Regression: Used in earlier versions for comparison and calibration

---

🧰 Libraries and Tools

• TensorFlow & Keras
• OpenCV
• NumPy
• PIL (Python Imaging Library)
• Matplotlib

---

🧪 Supported Object Classes

Includes common objects such as:

• person, car, truck, bicycle, motorbike
• traffic light, stop sign, bus, bench, dog, cat, laptop, cell phone, and many more (full COCO dataset class list)

---

🖼 Example Image Inference

from PIL import Image
import matplotlib.pyplot as plt

image = Image.open('/content/data/image.jpg')
detected = detect_image(image)
plt.imshow(detected)
plt.axis('off')
plt.show()

---

🎥 Example Video Inference

video_path = '/content/data/video1.mp4'
output_path = '/content/data/video1_detected.mp4'
detect_video(video_path, output_path)

The detected video will be saved with bounding boxes and labels overlaid.

---

⚙️ How It Works

1. Preprocessing: Resize images and normalize pixel values
2. Model Prediction: Run YOLO on the preprocessed image
3. Postprocessing:
   • Decode model outputs to bounding boxes
   • Apply Non-Maximum Suppression (NMS) to remove duplicates
   • Draw bounding boxes and class labels on image
4. Rendering: Use PIL and OpenCV to render final results

---

📁 File Structure

.
├── detect_image()       # Image object detection pipeline  
├── detect_video()       # Video object detection pipeline  
├── YOLO decoding logic  # Custom functions to decode YOLO output  
├── Bounding Box logic   # Includes NMS and IoU computations  
└── Draw utilities       # Renders boxes with color and labels  

---

⚡ Requirements

• Python 3.7+
• TensorFlow (tested on 2.x)
• Keras
• OpenCV
• NumPy
• PIL
• Matplotlib

💡 Make sure your environment supports GPU acceleration (CUDA/cuDNN) for optimal performance.

---

📦 Setup Instructions

git clone https://github.com/sudhanvad18/ObjectDetection.git
cd ObjectDetection
pip install -r requirements.txt

---

📌 Acknowledgements

• YOLO implementation based on the original YOLOv3 architecture
• COCO Dataset for object classes
• TensorFlow/Keras for deep learning tools

---

🛠 Future Work

• Switch to YOLOv8 or EfficientDet for better accuracy and speed
• Improve preprocessing for low-light or blurry inputs
• Add Flask/Streamlit UI for live video feed inference

---

📸 Example Output