Challenge 01 / Property / Retail / Visual AI

Smart Cities

How might AI turn visual data from retail and property spaces into actionable insights that improve customer flow, utilisation and operational response?

The operator of a busy property needs to know where people are moving freely, where they are waiting, and when an issue calls for action. A camera feed on its own leaves someone to watch the scene and interpret what is happening.

Working brief pending EIP confirmation

Scene 01 / Illustrative simulation

The model shows an entrance, checkout and marked through-route. In the second stage the checkout queue crosses the passage. In the third, a staffed second lane splits arrivals and the operator checks whether people can pass freely again.

A retail concourse where a growing queue begins to spill into the main walkway.

Preset scene key / Sample signals

Entry
Steady arrival flow
Checkout
Three people waiting
Passage
Through-route clear

An illustrative scenario

When a queue spills into the walkway

Imagine a mall concourse with a pop-up counter near an entrance. A short rush can look crowded without causing a problem. The operator needs to spot when the checkout queue starts to block the through-route and whether an intervention restores movement. The numbers below are invented for the walkthrough.

  1. Observe flow

    Start at the entrance and follow people past the checkout. The queue stays inside its zone, and visitors can still use the marked passage. Count arrivals, people waiting and people passing through as separate measures.

    Example signal
    A one-minute sample has 18 arrivals, three people waiting and no overlap with the passage.
    Operator choice
    Keep monitoring. Save this window as a comparison for the next interval.

  2. Spot a queue

    The checkout queue grows toward the entrance and crosses the through-route. Compare the outline of the queue with the passage boundary and see how long the overlap lasts.

    Example signal
    The example shows seven people waiting and a 20-second overlap with the passage.
    Operator choice
    Check the view and ask floor staff to confirm the obstruction before opening another lane. If the camera view is blocked, mark the estimate as uncertain.

  3. Respond

    The supervisor opens a staffed second lane. Arrivals split between counters, and the marked passage starts to clear. Follow both queues to see whether the blockage shifts toward the entrance.

    Example signal
    The passage is clear and two people are waiting at each counter. Recheck the entrance and both queues in the next interval.
    Operator choice
    Compare passage overlap and queue length before and after. Keep the second lane open only if movement improves without creating another bottleneck.

Design considerations

What a prototype could do

These are illustrative ideas for a prototype. EIP has yet to confirm datasets, integrations or judging criteria.

Who uses it

A property operations supervisor looks at the alert and decides whether to send staff. A floor team confirms what they see on-site. The operator needs a zone, a reason for the alert, and an indication of how recent and reliable the estimate is.

Example inputs

Use a synthetic or anonymised clip, a simple floor plan and timestamped counts by zone. A sample record might contain the 12:05–12:06 window, zone Checkout A, an estimated seven people waiting, 20 seconds of passage overlap and a recent frame timestamp.

Useful outputs

Show a floor plan with entry, queue and passage zones, a trend for queue length and a short alert explaining what changed. Keep an event history so an operator can compare conditions before and after opening the second lane.

What to measure

Hand-label a short sample and compare the estimated counts with the labels. Record missed and false queue alerts, time from obstruction to alert, and how long an operator takes to decide. Compare passage overlap before and after the intervention.

Test where the signal can mislead

  • A group stops to chat

    Check whether a brief stationary group looks like a queue. Use its position and duration to decide whether to ask for a review.

  • The camera view is blocked

    Mark the affected zone as uncertain and show when the last usable observation arrived. Show a missing count as unknown.

  • The second queue blocks the entrance

    Check both the entrance and the passage after staff intervene. A shorter checkout queue can still leave a worse route for arrivals.

Sample demo sequence

  1. Show ordinary flow

    Start with people passing the checkout and label the entrance, queue and through-route.

  2. Replay a queue spill

    Show the timestamp, estimated queue size and the passage overlap that led to the alert.

  3. Show the operator’s action

    Open the second lane in the example and compare what happens to both queues and the passage.

  4. Replay one failure case

    Show how the prototype flags uncertainty, and state what data it retains and why.