How to Run a Webcam Eye Tracking Study
Running a webcam eye tracking study means measuring where someone looks using a standard computer webcam rather than dedicated hardware. It can be conceptualized as a six-stage process: deciding what to measure, relating your stimuli to that measurement, choosing a calibration approach, preparing the environment and participant, collecting gaze data during the task, and analyzing the output against what the method can actually support. Labvanced's webcam eye tracking pipeline, the most rigorously benchmarked to date, reaches 1.2 to 1.4 visual degrees of accuracy against a research-grade EyeLink 1000 tracker (Kaduk et al., 2024), close enough to support most attention and visual-preference research questions without hardware, though not every research question.
None of these stages are unique to any single platform, though implementation details vary. Each section below covers the general methodological decision first, then points to how Labvanced implements it.
Table of Contents
The Webcam Eye Tracking Workflow
I. Decide What to Measure and Whether Webcam Eye Tracking Fits Your Question
What Webcam Eye Tracking Measures
A study built around webcam eye tracking typically records one or more of the following:
- Fixations: where and how long someone looks
- Saccades: rapid movements between fixation points
- Area of interest (AOI) analysis: whether and how long a participant looked at a defined region
- Gaze path: the sequence of visual attention across a stimulus
- Time to first fixation: how quickly a stimulus draws attention
Pupil size is technically extractable from some webcam pipelines, but considerably less reliable than with dedicated hardware, so it is rarely used as a primary outcome measure in a webcam study.
Whether the Method Fits Your Research Question
Whether the method fits a given research question comes down to the accuracy it can deliver. Webcam eye tracking is well suited to AOI and dwell-time analysis, fixation counts, free-viewing and visual preference paradigms, and general visual attention patterns. Design Considerations that Affect Webcam Eye Tracking Accuracy lays out the published accuracy figures behind that fit, and the design decisions that determine whether a given study reaches them, in full.
It is not well suited to sub-degree precision tasks, saccade dynamics, or letter-level fixation precision in dense text: the accuracy gap against a research-grade hardware tracker is real and worth designing around rather than assuming away.
II. Relate Your Stimuli to the Gaze Data You Need
Gaze tracking runs independently of what a stimulus actually is: an image, a block of text, a video frame, or a live interface can all be tracked the same way, since the method records where the eyes point relative to the screen, not what is rendered underneath. The design decision that matters is where regions of interest sit relative to each other and how much space separates them, since accuracy is not uniform across the screen and two closely spaced AOIs risk having a fixation near the boundary misassigned to the wrong one.
Labvanced's implementation: Presenting Stimuli in an Eye Tracking Task covers how any object type in the task editor becomes trackable without special configuration, and how to think about stimulus layout before touching a single eye tracking setting.
III. Choose a Calibration Approach
Any webcam eye tracking system estimates gaze position from a calibration mapping built by asking the participant to look at a series of on-screen points. Longer procedures with more points produce a more accurate mapping, but calibration is also the one part of the study where the participant has to actively engage with the system rather than simply attend to the task, so length is a real tradeoff against fatigue and dropout, not a setting to maximize by default. The right length depends on how tightly your AOIs are packed: a design with many small, closely spaced AOIs needs the accuracy a longer calibration buys; a design with a few large, well-separated regions can use a shorter one.
Calibration length sets the ceiling, but without a defined quality threshold, a participant whose calibration didn't converge is included in the dataset indistinguishable from one whose calibration did, rather than prompted to try again or excluded. Setting a maximum acceptable calibration error before data collection starts turns that into an explicit standard rather than an assumption.
Labvanced's implementation: How to Choose Eye Tracking Calibration Settings is a decision guide to calibration length, point count, quality thresholds, and the Physiology Toolbox Version your study needs.
IV. Prepare the Environment and the Participant
Webcam eye tracking depends on a face-tracking model reliably detecting facial and eye landmarks from the webcam input, so several environmental factors materially affect data quality before a single trial runs. Backlighting silhouettes a participant's face and reduces the contrast the model needs; a webcam positioned far from the screen adds error beyond what calibration corrects for; and glasses that cause reflections or carry a tinted or blue-light-filtering coating distort the light reaching the camera. None of these are unusual requirements, they are the same category of setup consideration a lab manages for a hardware eye tracker, applied to a setting where the researcher cannot physically check the participant.
Labvanced's implementation: The Participant Calibration Experience walks through the calibration flow participants see, including the practical lighting, glasses, and webcam-placement factors above and what happens when a participant drifts out of position.
V. Collect Gaze Data During the Task
Recording Gaze Data
Once calibration is complete, gaze coordinates are recorded and mapped against whatever regions of interest were defined at design time. Depending on the study design, recording can run continuously across the full task, or be restricted to specific areas of interest, capturing gaze only while a participant looks within a defined region rather than across the entire frame. Gaze is also rarely the only channel worth capturing: pairing it with other behavioral or physiological measures builds a fuller picture of a participant's response than gaze position alone.
Labvanced's implementation: Eye Tracking Data Output in Labvanced shows what a gaze export actually contains, including the per-sample confidence value that flags a blink or an undetected face. Using Shapes as Areas of Interest (AOI) covers defining the regions gaze is recorded against, using Polygon or SVG shapes as masks. Labvanced can also record complementary channels alongside eye tracking in the same task:
- Mouse tracking: cursor position and movement trajectories
- Emotion detection: facial expression classified into emotion labels
- Remote heart rate detection (rPPG): heart rate estimated from the webcam feed
Recalibrating During the Session
For longer or multi-block sessions, triggering recalibration at defined points rather than relying on a single calibration to hold for the entire session protects against position and lighting drift, a live methodological risk for any remote, unsupervised setup.
Labvanced's implementation: Setting Up Eye Tracking in a Task covers configuring recalibration cadence per task.
VI. Analyze the Output
Webcam gaze data is exported as a timeseries rather than one value per trial, since gaze position is recorded many times per second rather than once. Sampling rate is limited by how fast a browser can process webcam frames, commonly topping out somewhere between 30 and 60 Hz depending on the participant's hardware. From the raw gaze stream, the measures that support most study designs are fixation duration, number of visits to an AOI, time to first fixation, and dwell time, computed downstream in R, Python, or dedicated analysis software.
Labvanced's implementation: Recording and Exporting Gaze Data covers building the events that record gaze data and what the resulting export looks like once your study has run.
Summary Table
| Stage | Methodological decision | Labvanced implementation (see linked guide above) |
|---|---|---|
| I. What to measure | Choose measures the accuracy ceiling actually supports (AOI, fixations, dwell time) over ones it doesn't (saccade dynamics, sub-degree precision) | Accuracy and design considerations page |
| II. Stimuli and gaze | AOI spacing matters more than stimulus type | Presenting stimuli guide |
| III. Calibration | Length and point count trade off accuracy against fatigue and dropout, a defined quality threshold decides which calibrations are good enough to use | Calibration settings guide |
| IV. Environment and participant | Lighting, webcam placement, and eyewear affect landmark detection before any trial runs | Participant calibration experience guide |
| V. Data collection | Recalibrate at defined points to guard against drift in longer sessions | Task setup and gaze recording guides |
| VI. Analysis | Compute fixation duration, dwell time, and related measures from the timeseries export | Gaze recording and export guide |