Traffic Sign Detection in Autonomous Vehicles

Project Overview

This project, "Traffic Sign Detection in Autonomous Vehicles," focuses on developing a machine learning model to accurately identify traffic signs. The primary motivation is to reduce accidents caused by human error due to misinterpretation of traffic signs, especially in the context of autonomous vehicles. This initiative aims to contribute to safer roads by integrating advanced AI capabilities into automotive safety systems.

Motivation

Traffic accidents due to human error, particularly those involving misinterpretation of traffic signs, result in a significant number of fatalities annually. This project addresses this critical issue by leveraging Artificial Intelligence (AI) and Machine Learning (ML) to create a robust traffic sign detection system. As students of Electrical and Electronics Engineering with a specialization in AI & Cybernetics, the team aimed to apply their knowledge to a real-world problem, showcasing an AI-based control system's potential.

Objectives

The main objectives of this project include:

• Designing and implementing a machine learning model capable of detecting traffic signs with high accuracy.
• Utilizing Python libraries such as TensorFlow and Keras for model development.
• Contributing to the reduction of accidents caused by human error in autonomous vehicles.
• Developing an energy-efficient and robust convolutional neural network architecture for traffic sign identification.

Methodology

The project adopted a systematic methodology involving data collection, model training, and validation.

1. Data Acquisition: Gathering datasets from various online sources to extract useful parameters.
2. Model Training: Training a machine learning model (specifically a Convolutional Neural Network - CNN) on the collected datasets.
3. Iterative Refinement: An iterative process of training and testing, where the model was re-evaluated and refined if the accuracy did not meet the desired criteria. This ensured a high-accuracy model.
4. Validation: Final validation of the machine learning model to ensure its effectiveness.

The model's architecture consists of multiple convolutional and max-pooling layers, with dropout layers interspersed to prevent overfitting.

Technical Implementation

The core of this project involves a Convolutional Neural Network (CNN) built using Python, TensorFlow, and Keras.

Key Libraries Used:

• tensorflow
• keras
• matplotlib.pyplot
• numpy
• pandas
• cv2 (for webcam control and image pre-processing)

Model Architecture Highlights:

The CNN architecture is designed to be "thin yet deep," aiming for energy efficiency without requiring a Graphics Processing Unit (GPU) for training. The model layers include:

• Conv2D layers
• MaxPooling2D layers
• Dropout layers
• Flatten layer
• Dense layers

The model was trained and tested, showing favorable results in both loss and accuracy, indicating no significant underfitting or overfitting.

Project Files (Conceptual):

• main.py: Handles the machine learning model, including importing libraries, loading and pre-processing image data, splitting data, building the CNN model, and training it.
• cameracode.py: Responsible for loading the trained model (modelfinal.h5) and processing images captured from a webcam for real-time traffic sign detection.

Results and Discussion

The project successfully designed a model capable of detecting traffic signs with higher accuracy than initially anticipated. Both training and testing phases were successful, meeting the requirements of the problem statement. The model's performance was evaluated through loss and accuracy graphs, demonstrating its stability without underfitting or overfitting.

Conclusion and Future Scope

This project successfully demonstrated the feasibility of developing an accurate machine learning model for traffic sign detection. The successful training and testing validate the approach.

Future Enhancements:

• Expanded Training Data: Further training with a larger and more diverse set of traffic sign images.
• Object Detection: Expanding the model's capabilities to detect other road objects such as pedestrians, other vehicles (trucks, cars, buses), and even road lanes.
• ADAS Integration: Integration with Advanced Driver Assistance Systems (ADAS) for a more comprehensive autonomous driving solution.
• Challenging Environments: Adapting the model for more challenging road conditions, such as those found in countries like India.

Authors

• Jenish Murdia (20BAC10004)
• Shaun Jacob Varghese (20BAC10022)
• Varun Ram S (20BAC10038)

Under the Supervision of:

• Dr. Anirban Bhowmick (Assistant Professor, School of Electrical & Electronics Engineering, VIT Bhopal University)

---