Traffic Sign Recognition System

Overview

This project builds and compares two models for recognizing traffic signs from the GTSRB dataset:

• A Convolutional Neural Network (CNN) using TensorFlow/Keras.
• A Random Forest Classifier using scikit-learn.

Project Structure

TRAFFIC_SIGN_DECODER/
│
├── assets/
│   ├── Images/                       # Raw or test images
│   ├── cnn_model.keras               # Saved CNN model
│   ├── mapping.json                  # Custom class ID to name mapping
│   ├── Meta.csv                      # Metadata
│   ├── Test.csv                      # Test dataset metadata
│   ├── Train.csv                     # Training dataset metadata
│   ├── traffic_sign_rf_model.pkl     # Saved Random Forest model
│   ├── X_train.pkl / y_train.pkl     # RF training data
│   ├── X_test.pkl / y_test.pkl       # RF test data
│
├── dataset/                          # Full dataset (organized image folders)
│
├── source/
│   ├── CNN/
│   │   ├── data.py                  # Load & prepare data
│   │   ├── prediction.py            # Predict using CNN
│   │   ├── preprocess.py            # Preprocessing utilities
│   │   ├── test.py                  # Evaluate CNN model
│   │   └── train.py                 # Train CNN model
│   │
│   ├── RandomForest/
│   │   ├── data.py                  # Load & process data
│   │   ├── prediction.py            # Predict using Random Forest
│   │   ├── preprocessing.py         # Feature extraction (histograms, etc.)
│   │   ├── test.py                  # Evaluate RF model
│   │   └── train.py                 # Train RF model
│
├── trash/                            # Temporary or discarded files
│
├── .gitignore                        # Git ignore rules
├── README.md                         # Project documentation
├── requirements.txt                  # Project Requirements

📊 Data

The dataset used in this project is based on the German Traffic Sign Recognition Benchmark (GTSRB). It contains thousands of labeled images of traffic signs from 43 distinct classes.

🔢 Dataset Details

• Total Classes: 43
• Image Size: Resized to 64x64

🧾 Data Distribution

The graphs below illustrate the distribution of traffic sign classes in the dataset:

📈 Test Set Distribution

These charts help identify class imbalance, and guide decisions for augmentation or weighting.

Models

CNN Model

The implemented CNN architecture consists of:

• Convolutional layers with ReLU activation
• MaxPooling layers for downsampling
• Dropout layers to prevent overfitting
• Fully connected layers for classification

Random Forest Model

The Random Forest implementation uses:

• Ensemble of decision trees
• Feature importance analysis
• Hyperparameter optimization

Confusion Matrices

CNN Model Confusion Matrix

The confusion matrix for the CNN model shows the model's classification performance across all classes. The diagonal elements represent the number of correctly classified points, while off-diagonal elements represent misclassifications.

Random Forest Confusion Matrix

The confusion matrix for the Random Forest model:

Interpretation:

• The Random Forest model correctly classified 238 samples of Class 0, 221 samples of Class 1, and 261 samples of Class 2
• Similar misclassification patterns to the CNN, but with slightly different distribution

Training Data Performance

CNN Training Metrics

🧠 Models Overview

Model	Accuracy	Framework	Feature Extraction
CNN	~94.76%	TensorFlow	End-to-end
Random Forest	~89.68%	Scikit-learn	HOG (Histogram of Oriented Gradients)

---

Usage

Prerequisites

• Python 3.8+
• PyTorch 1.10+
• scikit-learn 1.0+
• pandas, numpy, matplotlib

Installation

# Clone the repository
git clone https://github.com/being-bilal/Traffic-Sign-Decoder.git
cd Traffic-Sign-Decoder

# Install dependencies
pip install -r requirements.txt