Webcam Eye Tracking for Reading Comprehension Support

ISEF Category: Behavioral and Social Sciences

Subcategory: Cognitive Psychology  ·  Difficulty: Intermediate  ·  Setup: Home Setup  ·  Time: 1 to 2 Months

The Hook

When you get stuck on a sentence, your eyes do not glide forward. They pause, jump back, and scan again. A webcam can pick up those patterns well enough to power a reading aid.

What Is It?

WebGazer.js is a browser tool that estimates where your eyes are on the screen using a normal webcam. Think of it like a moving map of your attention while you read. When the map shows longer fixations, more regressions, or repeated backtracking, the software can guess that a passage is getting hard.

A fixation is a short pause where your eyes stay on one spot. An adaptive prompt can say, “Read that sentence again,” or “Slow down here,” when the gaze pattern crosses a rule you set. Your project tests whether those prompts help comprehension, or whether they just interrupt readers.

Why This Is a Good Topic

This topic works well because you can measure it two ways, with gaze data and with quiz scores. It connects to online learning, accessibility, and tutoring, where a system that notices confusion early could support readers. You can learn experimental design, browser-based data collection, and basic statistics without a professional lab.

Research Questions

• How does passage difficulty change fixation duration in WebGazer.js readings?
• What is the effect of adaptive rereading prompts on comprehension quiz scores?
• Does prompt timing after a long fixation improve recall more than prompt timing after a set number of words?
• To what extent does prior familiarity with the topic change the accuracy of difficulty detection?
• Which gaze feature predicts missed questions better, regression count or average fixation length?
• How does the presence of prompts change reading time compared with a no-prompt condition?

Basic Materials

• Laptop with a working webcam.
• Modern browser such as Chrome or Edge.
• WebGazer.js or a similar browser gaze library.
• A set of high-school reading passages at matched reading levels.
• Multiple-choice comprehension questions for each passage.
• Google Forms or a similar survey tool.
• Notebook or spreadsheet for recording results.

Advanced Materials

• External eye tracker such as a Tobii device.
• Screen-recording software with timestamps.
• IRB or school research approval documents.
• Python or R for mixed-effects modeling.
• Annotated reading passages with difficulty labels.
• A quiet testing space with controlled lighting.
• Headphones or sound masking to reduce distraction.

Software & Tools

• WebGazer.js: Estimates gaze position from a regular webcam in the browser.
• Google Sheets: Organizes passage scores, prompt logs, and calibration notes.
• Python: Cleans gaze events and runs correlations or classification models.
• JASP: Runs t-tests, ANOVA, and effect size checks without paid software.
• RStudio Desktop: Fits mixed-effects models if you want a stronger analysis.

Experiment Steps

1. Define your reading signal, such as long fixations, backtracking, or repeated prompt triggers.
2. Choose passage sets that differ in difficulty but stay similar in topic, length, and format.
3. Plan a calibration check and a fallback rule for bad webcam tracking before each reading session.
4. Design the adaptive prompt rule, including the gaze threshold and the exact rereading instruction.
5. Decide your comparison groups, such as prompt versus no prompt or early prompt versus late prompt.
6. Set your outcome measures, including comprehension score, reading time, and gaze features, before you collect data.

Common Pitfalls

• Calibrating once at the start and never checking again, which lets gaze estimates drift as the reader moves.
• Mixing passage difficulty with topic interest, which makes it hard to tell whether confusion came from the text or the subject.
• Triggering prompts too often, which turns the reading task into a stop-and-start exercise and changes natural behavior.
• Judging the system only by quiz score, which hides whether the gaze rule actually detected difficulty at the right moment.
• Testing in changing light or with a tilted webcam, which weakens gaze estimates and adds noise to every result.

What Makes This Competitive

A stronger version of this project goes beyond a simple before-and-after score check. You could compare several prompt rules, test whether the model generalizes across passage types, and report how gaze errors affect the final outcome. If you pair clean controls with careful statistics and a clear failure analysis, the work starts to look like real research instead of a classroom demo.

Project Variations

• Test the same adaptive prompts on science passages, then compare them with narrative passages to see whether text type changes the gaze signal.
• Replace rereading prompts with summary prompts and compare which one helps comprehension more.
• Use mouse pauses or scroll pauses instead of webcam gaze and see whether simpler signals predict confusion nearly as well.

Learn More

• WebGazer.js GitHub repository: Read the project documentation and examples for browser-based gaze tracking.
• PubMed: Search review articles on eye tracking, reading comprehension, and cognitive load.
• NIH PubMed Central: Find free full-text papers on gaze behavior and adaptive learning.
• Institute of Education Sciences: Search reports on reading interventions and comprehension assessment in the What Works Clearinghouse and ERIC.
• MIT OpenCourseWare: Review free materials on statistics and experimental design for data analysis.