GTSRB Traffic Sign Recognition Project

Author: Rishab Hanamar
Git Link: https://github.com/rishab-14/Vehicle_Traffic_Sign_Detection_CNN_Metaclassifier

---

1. Project Overview

Project Title:
Traffic Sign Recognition Model

Problem Statement:
Accurate recognition of traffic signs is essential for road safety and autonomous driving systems. Real-world conditions such as varying illumination, weather, and sign appearance make this a challenging computer vision problem.

Objective:
To build and evaluate a CNN-based traffic sign recognition model using the GTSRB dataset that can classify images into 43 traffic sign categories with high accuracy.

---

2. Dataset

Dataset Name: German Traffic Sign Recognition Benchmark (GTSRB)

Source: https://www.kaggle.com/datasets/meowmeowmeowmeowmeow/gtsrb-german-traffic-sign

Number of Classes: 43

Image Size Used: 32 x 32

Class Distribution Handling:
Class imbalance was mitigated using shuffled and balanced mini-batches during training.

---

3. Project Structure

project_root/
│
├── contents/
│    └─gtsrb
│       ├── Train/
│       ├── Test/
│       └── Train/
│
├── traffic-sign-dectection-models.ipynb.ipynb
├── traffic-sign-detection-metalearner_main.ipynb
├── traffic_sign_detection_RQs.ipynb
│
├── models/
│    ├── complete_models
│    │    ├── custom_cnn
│    │    │    └── custom_cnn_model.h5
│    │    ├── meta_learner
│    │    │    └── meta_learner_model.h5
│    │    └── resnet50
│    │         └── resnet50_model.h5
│    └── weights    
│         ├── custom_cnn
│         │    └── custom_cnn_model_weights.weights.h5
│         ├── meta_learner
│         │    └── meta_learner_model.weights.h5
│         └── resnet50
│              └── resnet50_weights.weights.h5
│
├── outputs/
│   ├── Figures/ _<Contains figures from Rearch Questions>
│   ├── Tables/  <Contains tables from Rearch Questions>
│   └── <Contains useful pdf of models and dataset>
│
└── README.md

Model Architecture

4. Requirements

Python: 3.13.9

5. How to Run the Project

Step 1: Copy the data set from the link provide above (https://www.kaggle.com/datasets/meowmeowmeowmeowmeow/gtsrb-german-traffic-sign) and unzip it in the contents folder with the folder name "gtsrb". It should follow the above folder structure

Step 2: Update/Check the paths in all the notebooks. DATASET_PATH

Step 3:Run the notebook "traffic-sign-dectection-models.ipynb.ipynb" Here the two sepearte models for Custom CNN "custom_cnn_model.h5" and ResNet50 *resnet50_model.h5* are created. It is saved in the folder outputs in the completed_models directory

Step 4:Run the Meta Learner Notebook "traffic-sign-detection-metalearner_main.ipynb" which uses the models created in the step 1 and generates the complete model "meta_learner_model.weights.h5"

Step 5: Run the Notebook for All RQs "traffic_sign_detection_RQs.ipynb" which will create the required tables and figures which support each of the Research Questions

6. References

Refer to the report at the location "" for full details about the model. Architecture, ML Pipeline, Research Questions are all provided in it.