Welcome to the GitHub repository for the Traffic Sign Classifier, an innovative project developed as part of the DS3 Datathon. This initiative is a testament to our team's dedication to applying advanced data science techniques for enhancing road safety and the capabilities of autonomous driving systems.
The Traffic Sign Classifier is a sophisticated machine learning model designed to identify and categorize various traffic signs from images. Leveraging a comprehensive dataset, our approach integrates meticulous data preprocessing, exploratory data analysis (EDA), and the deployment of a state-of-the-art Convolutional Neural Network (CNN).
- Data Preprocessing: Detailed cleaning and normalization of image data to ensure optimal input quality for our CNN model. Exploratory Data Analysis (EDA): Thorough examination of the dataset to discern patterns and relationships, guiding our strategic modeling approach.
- CNN Architecture: Implementation of an advanced CNN, utilizing layers like batch normalization, dropout, and MaxPooling to enhance model accuracy and computational efficiency.
- Advanced Techniques: Incorporation of data augmentation to enrich training data, adaptive learning rates for improved training dynamics, and callbacks such as EarlyStopping and ReduceLROnPlateau to optimize training and prevent overfitting.
- Model Evaluation: Adoption of train-test split and cross-validation methods to rigorously evaluate model performance, ensuring its reliability and robustness.
Our Traffic Sign Classifier has demonstrated impressive accuracy in classifying traffic signs, showcasing the potential of machine learning in traffic management and autonomous vehicle technology. The detailed results and performance evaluations are available in our project documentation and Jupyter notebooks. Achieving a high accuracy score of 98% highest in the datathon kaggle competion.
Developed and worked on by the ABTV team
- Vaibhav Lakshmi
- Bahar Chidem
- Arina Azmi
- Tara Jorjani