Indivd
•
•
Intro
A shopping centre in Southern Europe recorded an average anonymous matching accuracy of 97.72% across 18 tested zones. Manual review checked whether Indivd correctly associated masked observations from different areas.
A shopping centre in Southern Europe recorded an average anonymous matching accuracy of 97.72% across 18 tested zones. The assessment checked how reliably Indivd matched observations of the same visitor between different areas.
This capability is known as re-identification, or ReID. It supports analysis of how visitors move between areas and which parts of a centre share visitors.
The assessment used samples collected on 16 September 2026. It covered 8,000 visitor movements, and compared automated matching decisions with manual review of masked image pairs.
Sixteen zones recorded accuracy of at least 95%. Two lower-performing zones were flagged for camera angle.
The centre
Understanding how areas connect
Footfall shows how many people enter an area. The companion shopping-centre counting QA report examines the accuracy of those counts. Understanding how different areas share visitors also requires reliable matching between observations.
Do visitors to one retail destination also visit another? Which areas share visitors? How do those patterns change after a campaign or layout change?
Re-identification provides the matching capability behind this analysis. An incorrect match can suggest a connection that did not happen. A missed match can leave a real connection unrecorded.
This assessment examined the accuracy of those matching decisions across the centre’s tested zones.
The check
What ReID means in this report
Re-identification is an established computer-vision term for matching observations of the same person across different cameras or views. Indivd uses the term for its anonymous visitor matching between zones.
ReID describes the matching capability. Indivd’s anonymization approach and the masking used during quality assurance are separate parts of how that capability is implemented and checked.
DEFINITION
Anonymous matching accuracy measures how often Indivd’s automated system agrees with manual review about whether two masked observations show the same person.
Comparing automated decisions with manual review
Reviewers examined pairs of masked images showing real visitor movements between two areas. For each pair, they decided whether the images showed the same person.
Indivd’s automated decision was then compared with the manual judgement, following its documented process for verifying data accuracy. This comparison was the verification method used to assess ReID accuracy.
The 18 published zone results averaged 97.72%. This is the arithmetic mean of the individual zone accuracies, with each zone given equal weight.
Results ranged from 88.06% to 100%. Sixteen zones achieved at least 95%. The remaining two recorded 88.06% and 88.14%, with camera angle identified as the reason for both lower results.
Reading accuracy alongside capture rate
Every tested zone lists a 60% capture rate.
Indivd defines capture rate as the proportion of visitors successfully re-identified across zone boundaries. Matching accuracy describes agreement with manual review for the evaluated matching decisions.
These measures answer different questions: how much movement the system captures, and how accurately it makes the assessed match decisions. The 97.72% average should therefore be read alongside the reported capture rate and individual zone results.
What the number carries
Evidence behind connections between areas
The assessment gives the centre a documented basis for evaluating the matching used in its movement analysis.
The average describes performance across the tested zones. Individual zone results show where accuracy differed, helping the operator assess connections involving the areas relevant to a particular decision.
The two lower results identify camera views to investigate. Any improvement following a camera adjustment would need a further assessment to verify it.
For planning, campaign evaluation and discussions about how areas share visitors, the useful evidence is the combination of matching accuracy, capture rate and performance by zone.
Common questions
What is ReID?
ReID stands for re-identification, an established computer-vision term for matching observations of the same person across cameras or views. Indivd uses this capability for anonymous visitor matching between zones.
What is anonymous visitor matching in a shopping centre?
Anonymous visitor matching associates observations from different areas with the same visitor without attaching a personal identity. It supports analysis of movement between zones, helping operators understand how visits to different parts of a centre connect.
What is cross-visitation in a shopping centre?
Cross-visitation describes the overlap between visitors to two defined areas. For example, it can show how often visitors observed in one retail area also visit another. The measure needs a clear time period, a defined visitor base and reliable matching between areas.
How is anonymous visitor matching accuracy measured?
Automated matching decisions are compared with manual review of masked image pairs. A reviewer determines whether each pair shows the same person, and the system’s decision is checked against that judgement. Results should identify the tested zones, sample and method used to calculate accuracy.
What is the difference between matching accuracy and capture rate?
Matching accuracy describes how often the evaluated matching decisions are correct. Capture rate describes the proportion of visitors successfully re-identified across zone boundaries. Both matter: high accuracy within the captured observations does not establish that every visitor movement was measured.
How can cross-visitation support shopping-centre planning?
It can help operators understand which areas share visitors and investigate whether campaigns or layout changes affect those connections. Matching quality and coverage should be considered alongside the observed patterns. Shared visits alone do not demonstrate that one area caused traffic or sales in another.
What should shopping-centre operators ask a visitor analytics provider?
Ask what the accuracy figure measures, how observations were selected for review, what proportion of movements is captured and how results vary by zone. Also ask how the system is verified after changes. A centre-wide average should come with enough detail to assess the areas relevant to your decisions.