Real-time anonymized crowd counting at Perform

[adam_gardner]

At the Developer Hub during Perform 2026 I ran a small experiment combining AI computer vision, Prometheus, and Dynatrace to see if we could reliably track how many people utilize a particular space (the developer hub). 

The setup was simple: a camera feeding an anonymized vision model, counts exposed in Prometheus format and an OpenTelemetry collector configured to scrape the metrics. Finally Dynatrace to visualize the metrics. 

It wasn’t flawless - the camera only ran for part of day one. Despite this, the main takeaway is clear: the concept works. With essentially zero manual effort we produced a realtime headcount for the developer hub. 

What we gained:

• Immediate operational insight: know when your booth or session is getting crowded. 

• Data-driven staffing: the potential to dynamically allocate staff based on real usage rather than guesswork. 

• Post-event analysis: understand peak times and traffic patterns. 

Privacy-first design 

We designed the system to be respectful and safe. All processing was anonymous: 

• The raw video feed was only used to track how many bodies were in frame at any one time. 

• The raw video was never displayed or stored in any way and we never tracked individuals. 

• The system only incremented counters: “a body walked in” or “a body walked out” - and never attempted identification. 

What this means for attendees 

Find the crowd-free route: Real-time counts let you avoid congested areas and pick the quickest path between sessions or booths, saving time and stress. 

Discover trending spots: See which talks, demos, or exhibitors are drawing attention right now so you can join the buzz (or intentionally skip it if you prefer quieter moments). 

Shorter lines, more value: Live occupancy data helps you choose booths with shorter waits, so you spend less time standing and more time having meaningful conversations. 

Better accessibility and comfort: Attendees with mobility needs or sensory sensitivities can plan visits when spaces are less crowded, making the event more welcoming for everyone. 

Personalized timing: Use popularity feeds to time visits to keynote sessions or workshops when space and engagement are optimal for your goals. 

Safer, calmer experience: Real-time alerts about overcrowding let you avoid bottlenecks and reduce the risk of uncomfortable or unsafe situations. 

Make the most of your day: With an at-a-glance view of crowd flow and hotspots, you can prioritize what matters — networking, learning, or demos — without wasting time on dead-ends. 

All of this while keeping privacy front and center: the system only reports anonymized in/out events, so attendees get a smoother, more efficient conference experience without being tracked. 

Where we could go next 

This proof of concept opens easy extensions for next year: 

• Realtime popularity feeds for each expo tower or booth. 

• A heatmap of the whole expo floor to visualize flow and dwell time. 

• Threshold alerts and automated staff notifications when areas get busy. 

• Aggregated analytics for better layout planning and future promotional timing. 

Final thoughts 

Small experiments like this are how events get easier to navigate. For attendees, the payoff is immediate: less time hunting for popular sessions, fewer waiting times / lines, and more control over your day. By surfacing anonymized, real-time occupancy data, we can help you find quieter moments for focused conversations, spot the liveliest demos worth your time, and avoid congested routes. I’m excited for what this PoC could mean for Perform 2027 at the Venetian – see you there!