Development and testing of software for automatic measurement of blink behavior
Jorge Otero-Millan, Professor
Optometry
Applications for Fall 2026 are closed for this project.
Blinks are usually treated as noise in eye-tracking research: intervals to be detected and discarded before analyzing gaze. Yet blink rate, duration, and timing relative to task events are informative measures of visual and cognitive state in their own right, and they are also relevant to ocular surface health. Current detection methods rely on ad hoc heuristics — signal dropout, pupil-size thresholds — that are tuned per device, handle partial blinks poorly, and are rarely validated against ground truth. This project develops and rigorously tests open-source software that detects blinks from eye videos and characterizes them (onset, offset, closure duration, amplitude, partial vs. complete). The student will help build a hand-annotated reference dataset, implement and compare candidate algorithms and quantify accuracy across recording conditions and eye trackers.
Role: - Manually annotate blink onsets and offsets in eye videos to build a ground-truth dataset, using and helping refine a lab annotation tool
- Implement and document blink-detection algorithms in Python (or MATLAB), starting from existing lab code
- Design and run quantitative benchmarks: precision/recall of blink detection, error in estimated blink duration, sensitivity to frame rate, illumination, and eyelid occlusion
- Test generalization across recording setups (different cameras, eye trackers, and participants) and report where each algorithm fails and why
- Maintain code under version control (Git/GitHub) with unit tests, example data, and usage documentation
- Read papers on blink detection and blink physiology to inform algorithm design
- Present progress at weekly lab meetings and contribute a methods section to a resulting paper or software release
Qualifications: - Comfortable with software development testing.
- Comfortable with AI assisted software development and testing.
- Background in signal and/or image processing, statistics, or computer vision.
Day-to-day supervisor for this project: Meng Lin
Hours: to be negotiated