AI Glossary / Computer vision and multimodal AI / Advanced
Scanpath
Definition
A scanpath is an ordered description of where a person's gaze moves while viewing a stimulus, often represented as a sequence of fixations linked by saccades. Depending on the representation, it can include positions, durations and timestamps. A measured scanpath reflects both eye-tracking observations and the processing choices used to identify gaze events.
What it means
A heatmap can show where looking was concentrated, but a scanpath preserves information about sequence: what was inspected first, where the gaze moved next and which areas were revisited. This makes it useful for studying visual search and interaction. It does not directly reveal why someone looked at a location or whether they understood it. The same sequence can arise from different goals, levels of familiarity or difficulties, so interpretation needs the viewing task and other evidence.
How it works
An eye tracker records gaze-related measurements over time. Processing can classify samples into events such as fixations and saccades, then map them onto the displayed stimulus. A gaze plot commonly uses numbered circles for fixation order, circle size for duration and connecting lines for transitions. Raw gaze traces and event-based scanpaths are different representations. Calibration quality, missing data, sampling and event-detection settings affect the resulting sequence. For moving or scrolling content, coordinates must also be aligned with the relevant stimulus state rather than treated as a fixed image throughout.
Three practical examples
- A usability researcher examines whether participants move from a page heading to the instructions before attempting a task.
- A reading study compares ordered fixations and revisits while retaining the timing and text position of each event.
- A training team explores how experienced and novice observers inspect a diagram, without assuming that either group's sequence proves comprehension.
Think of it this way
Imagine a travel diary listing the places someone visited in order, with the time spent at each stop. It contains information that a map of total visits would lose, such as returning to an earlier destination. Yet the itinerary does not reveal the traveler's motives. A scanpath similarly records a viewing sequence, while the reasons for that sequence require context rather than being visible directly in the line connecting the points.
Worked example
In an original simplified sequence, a participant fixates region A for 200 milliseconds, region B for 300 milliseconds and region A again for 100 milliseconds. The ordered region sequence is A→B→A, with 600 milliseconds of total fixation duration. Region A accumulates 300 milliseconds, equal to region B, but it receives two visits rather than one. A duration-only summary would hide that difference. The 600 milliseconds excludes transition times, blinks and any unclassified intervals, so it should not be reported as the total elapsed viewing time without additional timing information.
Illustrative case study
A fictional software team tests a setup screen. Several participants look from a configuration field back to its label repeatedly before continuing. The scanpaths suggest a region worth investigating, but interviews reveal different explanations: some users are checking terminology, while others are verifying an entered value. The team revises unclear wording and evaluates task success in another test. This scenario uses gaze sequence to guide investigation. It does not treat repeated looking as definitive proof of confusion or assume that fewer fixations necessarily indicate a better interface.
Common misconceptions and mistakes
A scanpath is not a direct transcript of thought. Long fixations can have several explanations, and missing fixations do not prove that information was entirely unprocessed. Another mistake is combining participants into one sequence and presenting it as a typical individual's actual behavior. Event-detection settings can change counts and durations, while mapping errors can place gaze on the wrong interface element. Connecting lines in a plot also simplify movement between events; they should not automatically be interpreted as exact continuous trajectories through every intervening location.
Business use
Scanpath analysis can help teams investigate navigation, instructions and visual search in products or training materials. Start with a specific task and measurable outcome, then use gaze sequence alongside accuracy, completion time and participant feedback. Preserve individual variation when comparing groups, and document event detection and stimulus alignment. Avoid turning one striking gaze plot into a broad claim about all users. The practical value comes from identifying testable design questions and evaluating improvements, not from treating a visually compelling trail as a complete explanation of user behavior.
People also ask
How does a scanpath differ from a heatmap?
A scanpath retains sequence information that a static aggregate heatmap generally does not.
Are fixations directly recorded as finished events?
They are commonly identified from measurements using processing rules.
Does revisiting an area always mean confusion?
No. It can reflect checking, comparison or other task-related behavior.
Check your understanding
Which region had more visits in the example?
Answer: Region A, with two.
Which had more accumulated fixation time?
Answer: Neither; both had 300 milliseconds.
Why was 600 milliseconds not necessarily the full viewing interval?
Answer: It summed only the stated fixation durations and omitted transitions and other possible intervals.
Related terms
- Saliency
- Fixation
- saccade
- gaze plot
- area of interest
- eye-tracking calibration
Learn next
Write two different fixation sequences that produce the same duration totals by region. Compare their revisit patterns, then identify what a heatmap would preserve and lose. Finally, state which additional task observations would help interpret why the sequences differed.
More in Computer vision and multimodal AI
- Saturation (point operation) - A saturation point operation changes the color intensity of each pixel relative to a neutral gray reference, using only that pixel's input values.
- Scale space - Scale space is a framework for representing an image or signal at multiple levels of spatial detail.
- Scale-space axioms - Scale-space axioms are mathematical requirements used to define what a family of progressively coarser image representations should do.
- Scatter matrix - A scatter matrix summarizes multivariate variation through sums of outer products of centered observation vectors.
- Scene text
- Seam carving - Seam carving is a content-aware image-resizing technique that removes or inserts connected paths of pixels called seams.
Sources
Tobii, Visualizations: https://developer.tobii.com/xr/learn/analytics/fundamentals/visualizations/ Scanpath Studio, Scanpath visualization: https://lacclab.github.io/scanpath-studio/guides/scanpath-visualization/ The A→B→A calculation, travel-diary analogy and setup-screen case are original educational illustrations. They distinguish measured viewing sequence from inferred motives and do not diagnose a participant's mental state.
Previous: Scan Operation Next: Scatter matrix