How Much Is Video Analytics Software Worth?

We are looking for figures about the value of embedded vision software. For instance, in cameras/DVRs/NDRs: what's could be the value of software versus hardware? Any insight will help:).


Software only sales are clearly very low. You have companies like Agent VI, BRS Labs, OV (their software side, not patents) and none of them are doing much revenue (most certainly, all under $10 million USD). There are dozens of small software analytics companies around the world but I doubt the average one does more than a few million a year.

On the hardware side, it's not much better, though that's where people try to fudge the numbers, i.e., if Sony sells a camera with 'analytics' built in, count the whole cost of the camera as an analytic sales. Here's an absurd 2009 billion dollar analytics projection that obviously never came true.

Overall, there's only a handful of embedded/hardware offerings that depend on analytics - VideoIQ and 3VR being two notable ones. I would have said Cernium/Archerfish but they are gone now.

If you say the whole video surveillance market is somewhere between $10 and $20 Billion, the portion that comes from video analytics is no more than 2 or 3% - say $200 to $600 million.

Hello John. Tx for your comments and for your website: I discovered it a month ago and got already tons of very valuable information:).

Here is more info about what we would like to price. Our University has developed a change detection algorithm that seems to have quite some benefits in terms of speed, precision, and "universality" of use.

It seems that cameras/DVRs/NVRs might be an obvious market to us for our change detection algorithm. And of course, the question is how much we should charge for it. The answer is definitely not an easy one as we are dealing with many markets, many applications, ... Especially as common alternatives (MoG) are free of charge. That's why we would like to use benchmarks from other basic packages embedded in chips to set our price point.

We cannot of course be compared with VMS', VCA s/w, etc. as they deliver much more than what we do. But they could actually also be interested in our algorithm and the figures you mention here above are definitely valuable for our research.

Regards to all,

Pascal

Pascal, I know at least a dozen university CV projects have tried to commercialize their research into commercial video analytics. Typically, it's challenging.

I have seen two models:

  • Deliver full systems typically to local large security users - The university team has internal systems / field engineer(s) who go out and setup the software, servers, integration with existing system.
  • License to a manufacturer, get paid a small royalty per channel and let the manufacturer deal with integrating, marketing, etc.

The former approach (full systems) does not scale well and is expensive for users. The latter approach is tough because manufacturers typically want to pay very little as the market is not that hot, there are underlying patent litigation issues and there are lots of universities with other offerings.

I know this sounds pretty negative all in all, I just don't know how to realistically make such a plan work.

Hi John. isn't any start-up a challenge? ;). We realise all this and have to cope with that. I am an entrepreneur and have several start ups behind me so I am fully aware of the difficulties paving the way.

But in this case, my question is really about pricing, not about the business model (although it is a good question too;)). I know we won't get a "x,xx$" answer, but as i said: any insight would help.

Selling analytics to end users, you might be able to get $100 to $1,000 per channel.

Selling analytics to manufacturers, you might be $3 to $50.

The difference obviously is that manufacturers will buy in bulk and will handle all customer support, marketing, integration, etc.

This is a good guideline, but I think it also depends on what the analytics can DO in terms of providing value.

Some examples:

"Perimeter Protection" systems can usually show a short-term ROI and provide more measureable results, so they can get a good price.

"Storage Optimization" analytics can also provide measurable results because you can intelligently manage storage systems to reduce how many disks you need.

"Business Intelligence" analytics can vary more widely in price. Just how valuable is a report of people coming through the front door of a Home Depot? Sometimes the value here is as much about the presentation of the tabulated data as the actual analytics.

Mixed in with that is hardware/processing requirements for your algorithms. Something that can run with minimal overhead is easier to embed in lots of devices. If you tell us it needs a 1U server dedicated to processing a couple of low-res streams, it's probably a non-starter.

I was trying to avoid that question becomes obviously it depends on what it does, how well it does it and how comparatively lacking it is in the market. That's also why my range was so broad.

That said, those are obviously critical when it comes to more specific pricing and to potential viability.

Pascal, can you share with us what is new with your potential analytics offering, and what application and market you are targeting? These are key factors in triangulating on a potential price point, but I agree with the general pricing framed by John. The security market is a tough one to enter and for a start up it's tough to justify a business plan around analytics because there is not a very clear ROI. But it's not impossible, and there are many non-security analtyic applications and markets that are very interesting.

@ Brian & Skip.

I didn't want to "market" our offering here but based on your questions, here some info about what we believe we can offer.

Our University has developed a change detection algorithm (see vibeinmotion.com for references, etc.). It performs only and only change detection - pixel by pixel. This is what it does and it is absolutely not our goal (due to resource constraints) today to develop verticals around our algorithm. So our targets are clearly "basic" devices that can easily integrate such a simple s/w packet OR VCA companies that want to use us as brick in their solution. We are totally aware that we won't be able to extract as much value by doing so, but we have no other choice.

Now, the question we have here is how much value we can extract versus for instance a MoG that is free. The answer lies of course with the extra benefits we bring. Again, my point is not to market our product here, but here is what we think we can bring to the table:

- ViBE is at least 10 times faster than a MoG. This is critical when treating large volumes of data (more images/channels or higher quality). It is thus also more friendly in terms of computing resources. In commidity markets such as cameras, DVRs, etc. this might help lower overall costs of components.

- It is also more precise: testers have told us that this increase in precision seems to end up in +/- 50% better classification of true postiives/negatives. This can result in lower storage capabilities, faster analysis, and better results in critical situations.

- ViBE is "universal" in that it requires no critical hypotheses: its model is based on a set of past samples. This should deliver some benefits in easier set-ups and limited maintenance costs.

We realize that those gains will of course on the end application.

Regards to all. Pascal