Multiple VA Applications On Camera

I have a question regarding camera-side video analytics: is it possible to run multiple VA-applications simultaneously on a single camera? E.g. motion detection, loitering, running, left luggage, crowds/queues, tailgating, face recognition, obscured faces etc.. If so, can the applications be from different suppliers?


Camera side? Theoretically possible, often not possible due to resource constraints.

Are you using Axis or? What can be run depends on the camera manufacturer (i.e., do they support the app) and how much processing it has (which can vary based on the camera model) to do the analysis?

Generally, though you can do basic VMD + one other analytic. The problem gets more likely when doing combination of face recognition + tripwire, etc.

Also, you'll need to make sure that the camera FoV / angle is appropriate for both since analytic applications vary in what type of scene they need (e.g., face recognition tends to need the opposite camera positioning as people counting).

If you can provide some more details about what specific cameras and providers you are looking at, we can be more precise.

Many thanks for your prompt reply. We haven't specific cameras or providers in mind yet. We are in the process of deciding whether we should have camera-side or server-side analytics (or both?), and I was wondering whether this (multiple applications in parallel) might be a factor to take into account. Suppose we want to do multiple analytics on a single stream, could that be a reason to do the analytics server-side instead of camera-side? Or to do some analytics camera-side and other analytics server-side?

Multiple applications simultaneously definitely needs to be taken into account but that would not be the first thing I decide.

(1) What does the business / operation need? (obviously)

(2) Who has those types of analytics that actually work well for your application / scenarios? (should not be taken for granted)

(3) Once I find analytics that work appropriately for my needs, I'd then start figuring out the logistics. What cameras can this run on? How much load does it take, etc.?

Technically, yes, it's quite possible. Realistically, it depends on a number of variables.

To answer the last part of your question first, it's unlikely that you would be able to run VA applications from different developers on the same camera at all. And even if you *could*, I would personally avoid it, as the risks and troubleshooting difficulties would likely make you regret the decision.

On our cameras (Avigilon), we currently focus on object-classification analytics, where the system can identify (eg: classify, not identify as in saying "This person is Bob") people and vehicles in motion in the scene.

The rough logic flow in our camera is that objects are classified BEFORE any rule consideration, we attempt to classify any moving object as a person or vehicle (or, ignore it if it doesn't look like what we would expect, eg: a dog, or a shadow). Once the object is classified, the rules engine is consulted to say "is this object breaking one or more rules?".

This approach allows you to run multiple kinds of analytics rules simultaneously, because the hard part (object classification) is handled separately from the rules part. You can have a tripwire in one part of the of the field of view, a loitering rule in another, and an object present in region in another area. However, it's not very common for people to run multiple rules types, they usually just end up using "object present in region", as that tends to work best overall for the standard security application.

For any analytics scenario, you need to think more about the FOV and camera placement/coverage to give the best possible view to the analytics algorithm. For this reason, a camera deployed for face rec is going to have a much different FOV than a camera deployed for loitering (a more straight-on shot, usually at a choke point area vs. a higher up overhead shot).

Given the wide variety of analytics options available in the market (in terms of performance, practical coverage area, resolution, etc.) I would personally recommend not trying to find one "do it all" solution right now, but instead prioritize your needs and find the right tool for each job. Pay attention to what it takes to setup/tune the performance and to maintain that performance over time. Investigate integration options with your VMS of choice and how alarms are brought into the VMS, automation flexibility, and so on. Also, be sure that you can do a real-world test of the performance in YOUR environment, don't rely just on demo videos or spec sheets. Pay attention to daytime vs. night time performance and coverage area/range. Test false alarm immunity by not just concentrating on "does it catch a person" but also "does it ignore the proper things".

Samsung cameras do allow you to run multiple analytics at once.

  • There are some analytics built in, which can all run at the same time.
  • You can also add Open Platform apps, which can run at the same time as the built in apps.

You can also run multiple Open Platform apps at the same time, as long as they don't max out the CPU and memory of the camera. Something like this would be recommended to be tested before purchase. Some apps use more resources than others, and would not let another app run concurrently.

Absolutely,

We define multiple loitering areas, virtual fence crossings, and these VA events are used to power up military LED's spots mounted on PTZ's to place light, and Point / ZOOM cameras to the VA event.