Nisha Goswami

Challenge

Drowsy driving is a major cause of road accidents, and there was a need for an accurate, real-time detection system that could alert drivers before fatigue leads to dangerous situations.

Solution

Developed a YOLO-based computer vision model in PyTorch to detect signs of driver drowsiness by monitoring facial landmarks such as eye closure and yawning frequency. The system processes live video feed from a camera and triggers an audible alert when drowsiness is detected.

Impact

• Achieved over 90% detection accuracy in controlled test scenarios.
• Reduced reaction time to driver fatigue by providing immediate visual and audio alerts.
• Designed a modular architecture that can be integrated into vehicle infotainment or safety systems.

Challenge

There was no lightweight and accurate way to recognize celebrity images via a web interface, especially using scraped data without huge datasets.

Solution

Built a pipeline to scrape celebrity images via Google Images, processed them using OpenCV and NumPy, trained an ensemble of models—including XGBoost, Random Forest, LightGBM, and Logistic Regression—and deployed the classifier through a Flask-based web UI.

Impact

• Achieved reliable recognition across five popular personalities with limited dataset size (~900 images).
• Delivered a responsive Flask app for real-time image classification via drag-and-drop interface.
• Demonstrated how to combine web scraping, feature engineering, and ensemble learning into a cohesive, user-facing project.