Tracking 15-20 people — what performance issues will I hit?

[scottbraddon]

Hey,

I’m new here so apologies if this post lacks nuance but please let me know if you need clarification.

I’m looking at using 6-8 ZED 2i cameras to track 15-20 people in a room at the same time. Planning to use Fusion in Network mode with separate machines per camera pair, HUMAN_BODY_FAST, C++ or Python.

Has anyone pushed the system with this many people? What performance issues did you run into?

A few things I’m wondering about from older forum posts:

• Does tracking still slow down per person on SDK 5.x?
• Does Fusion itself become a bottleneck with more cameras?
• Any issues running for long sessions (60-90 min)?
• How does it handle colored lighting (LEDs, not white light)?

Appreciate any real-world experience. Trying to figure out if this is realistic before buying hardware.

Thanks

[Myzhar]

Hi @scottbraddon

@scottbraddon

Has anyone pushed the system with this many people? What performance issues did you run into?

It’s not possible to reply to this question since the performance results depend on the available computing power of the single hosts and the fusion server.

@scottbraddon

Does tracking still slow down per person on SDK 5.x?

What do you mean?

@scottbraddon

Does Fusion itself become a bottleneck with more cameras?

This depends on the quality of the network connection and the computing power of the fusion server.

@scottbraddon

Any issues running for long sessions (60-90 min)?

If the network connection is good, no issues are expected.

@scottbraddon

How does it handle colored lighting (LEDs, not white light)?

As long as the cameras are working in visible light conditions, the people detection module can work flawlessly.
If colors change dynamically, you could expect random tracking errors in the case of occlusions.