How is the accuracy of entry counts evaluated?
Aura Vision runs an accuracy audit on every location during setup, then randomised audits every quarter — each one a manual ground-truth comparison against video samples.
Aura Vision runs an accuracy audit on every location during setup, then randomised audits every quarter. Each audit is a manual ground-truth comparison: a human counts people in a sample of video clips and the result is compared to what the AI counted in the same period.
What the audit checks
- Entry-count accuracy — does the AI’s number match the ground-truth human count?
- Staff vs. customer split — is the AI correctly excluding employees?
- Demographic accuracy — does age and gender estimation match the human label?
The accuracy figure reported for your store comes directly from these audits.
When an audit flags an issue
When an audit flags a count discrepancy, we work with you to resolve the cause — typically one of:
- Obstructed view. A product display or point-of-sale has moved into the camera’s line of sight. The fix is moving the obstruction or repositioning the affected display.
- Sub-optimal camera angle. The current camera view doesn’t fully cover the entrance pattern. The fix is switching to an alternative camera with better coverage.
- Camera repositioned. A camera has been moved since the last audit. We re-align the entry line and re-audit.
In most cases the fix is a small adjustment and accuracy is restored in days.
Quarterly audits
Every store is randomly sampled for re-audit each quarter. The sample includes peak hours and off-peak hours so the result reflects accuracy under realistic conditions, not just quiet periods.