🛑 Stop Sign Detection and Distance Estimation using Stereo Vision and YOLO

📌 Overview

This project implements a robust and efficient system for real-time stop sign detection and distance estimation using a combination of deep learning (YOLO) and classical stereo vision techniques.

Developed with affordability and high performance in mind, this solution is ideal for applications in autonomous driving, advanced driver-assistance systems (ADAS), and robotics — where accurate environmental perception is critical.

Unlike traditional object detection systems that only identify objects, this project leverages stereoscopic triangulation to provide precise depth information, giving vehicles or robots a crucial understanding of their surroundings in 3D space.

---

🚀 Key Features & Advantages

• Hybrid Approach for Enhanced Perception
  Combines YOLO for real-time object detection with Stereo Vision for precise distance estimation.

• High Detection Accuracy
  The YOLO model achieves over 80% accuracy in detecting stop signs, even in complex scenes.

• Accurate Distance Estimation
  Uses the Stereo SGBM (Semi-Global Block Matching) algorithm to compute disparity maps and apply the triangulation formula:
  Z = (f × B) / d

• Cost-Effective Alternative
  Requires only stereo cameras — making it significantly cheaper than LIDAR or RADAR.

• Rich Visual Data
  Unlike LIDAR or RADAR, stereo cameras offer full-color and texture-rich images, enabling deeper scene understanding.

• Real-Time Performance
  Efficient YOLO inference + optimized stereo algorithms enable fast processing.

• Modular and Maintainable Codebase
  Built around the StereoYOLOProcessor class for clarity, modularity, and ease of future updates.

---

⚙️ Technical Details

🔍 Models and Algorithms

• YOLO (You Only Look Once) – Fast and accurate object detection for bounding boxes and labels.
• Stereo SGBM (Semi-Global Block Matching) – Computes pixel-wise disparity between stereo pairs.
• Triangulation Formula – Converts disparity to distance:
  Z = (focal_length × baseline) / disparity

📸 How It Works

1. Stereo Image Capture – Left and right frames are captured from a calibrated stereo camera pair.
2. Image Rectification – Aligns images along epipolar lines to prepare for matching.
3. Stop Sign Detection (YOLO) – YOLO processes one image (typically left) to detect and localize stop signs.
4. Disparity Map Generation (SGBM) – Disparity computed from stereo image pair using SGBM.
5. Distance Estimation – The average disparity inside the bounding box is used to calculate the distance.

---

📁 Project Structure

STOP_SIGN_PYTHON/
├── .pycache__/
├── .venv/                      # Python virtual environment
├── extracted_tesla_frames/     # Processed or raw frame data
├── STOP_LEFT_frames/           # Left image frames
├── STOP_RIGHT_frames/          # Right image frames
├── scene0/
├── scene1/
├── video/
│   ├── LEFT.mp4
│   └── RIGHT.mp4
├── config.txt                  # Configuration (model paths, thresholds, etc.)
├── detection.py                # Main processing script (detection + depth)
├── last.pt                     # YOLO model weights
├── packages.py                 # Custom modules
├── README.md                   # This file
└── video.py                    # Video capture/processing
.gitattributes
.gitignore

Note: The main class StereoYOLOProcessor is likely defined in detection.py.

---

🛠️ Getting Started

1️⃣ Prerequisites

Make sure the following are installed:

• Python 3.8+
• Git

2️⃣ Clone the Repository

git clone https://github.com/HodayaMoyal/Streo_Stop_Sign.git
cd STOP_SIGN_PYTHON

3️⃣ Create a Virtual Environment

python -m venv .venv

4️⃣ Activate the Environment

Windows:

.\.venv\Scripts\activate

macOS/Linux:

source ./.venv/bin/activate

5️⃣ Install Dependencies

If you have a requirements.txt:

pip install -r requirements.txt

Otherwise, install manually:

pip install opencv-python numpy ultralytics configparser

6️⃣ Download Models and Calibration Files

Ensure the following:

• last.pt (YOLO weights) is in the project root.
• camera_calibration.yml is available (with intrinsic/extrinsic params).

If not calibrated yet — see the Calibration section.

7️⃣ Configure config.txt

Update settings such as:

model_path=last.pt
confidence_threshold=0.5
calibration_file=camera_calibration.yml
num_disparities=64
block_size=11

8️⃣ Run the Project

Run the main script:

python detection.py

The script will output detected stop signs, bounding boxes, and estimated distances. You can optionally modify it to save output frames.

---

🧪 Calibration

Stereo calibration is essential. Use OpenCV’s calibration tools or a dedicated script (e.g., calibrate.py) to generate camera_calibration.yml.

---

🤝 Contributing

We welcome contributions!

# Fork and clone
git checkout -b feature/my-feature
# Make your changes
git commit -m "Add new feature"
git push origin feature/my-feature
# Open a Pull Request

---

📄 License

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