This project focuses on implementing a real-time helmet detection system using the YOLO v8 model. The researchers utilized two image datasets sourced from Kaggle, both containing annotated images specifically for helmet detection. These datasets primarily facilitate the binary classification of helmet presence, categorizing images into "With helmet" and "Without helmet" classes.

Dataset Insights

During the exploration phase, it was observed that the datasets exhibit data imbalance, showcasing varying counts between images depicting helmets and those without helmets. Recognizing this imbalance is crucial as it may impact the YOLO v8 model's performance during training. Understanding these key characteristics of the datasets forms the cornerstone of our approach in subsequent phases. This understanding enables us to address challenges related to class imbalance with the use of data augmentation techniques and ensures the robust training of the model for effective real-time helmet detection within the domain of traffic monitoring and surveillance.

Contributors

Ken C. Aquitan

Christian A. Muaña

Dawn Angela R. Velasquez