Video Analytics And Low PPF Impact?

I have a camera that has trip wire analytics, however the fenceline I want to monitor will come in around 10-15 PPF. Given distance and variable lighting conditions I am not confident the camera analytics will work. Any opinions?

I am inclined to use a photoelectric detector solution for the fenceline.


Without more details on the camera and analytics packages, it's hard to say. I don't generally spec things in a ppf or pixels on target manner (because most people don't think about the designs that way), but if I were, my minimum recommendation would be about 8ppf for our HD cameras. So, your 10-15ppf application is theoretically acheivable with at least *some* analytics (mine, ;)

A photoelectric detector seems like a false alarm generator to me, but I'm not sure about your budget or the site environment.

When you say "variable" lighting conditions, do you have an estimate of the lowest expected lighting levels at the target area (fenceline) and a rough idea of camera to target distance/range?

Brian, we do use VideoIQ cameras on a routine basis. (however harddrive failures have been a concern - especially when the site is a 4 hour drive to replace)

Seconded on the question about camera and analytics package.

At my (fairly ignorant) level of understanding, I would have thougth that video analytics would have poorer false alarm/missed detection Receiver Operating Characteristics as compared with a photoelectric detector tripwire.

At retail entry/exit points, photoelectric detectors seem very reliable, generating no false alarms while providing positive response to every entry or exit. Of course, they cover a relatively short distance, in indoor lighting, at a normally closed door. Garage door photoelectric sensors don't seem to prevent closing except when something is in the way or when one end of the sensor pair is knocked awry.

I would have thought that the challenge with a perimeter photoelectric tripwire would be weather exposure and cable runs, while the video tripwire, while inferior in performance, should be a simpler, weather-protected, and more tamper-proof installation.

I've rambled a bit but what I'm really asking is, could you help clear up my misconceptions please?

Thanks.

I have gotten a quote from Optex on a wireless photoelectric solution, batteries on units good for 4-5 years and alarm on degradation. Affordable and reliable.

I am inclined to pursue that for perimeter compromise. Video as a supplement. I hate to put my name on the line for a video analytic solution that wont be reliable? The sales engineer from the vendor is not certain it will work without testing.

We would certainly be viewing at better than 8 PPF on fenceline. 3MP camera, 3-9mm focal length. Which is the best camera resolution this vendor currently has with trip wire analytics. (actually they have a 5MP box that may do trip wire)

"The sales engineer from the vendor is not certain it will work without testing."

That basically means it won't work :) Probably a political way of expressing that.

Going to higher resolution (especially 5MP) has its own drawbacks. Despite the higher pixels per foot (though only modestly so over a 3MP - ~25%), typically it has worse WDR performance and low light (especially increased noise which is a real problem).

RJ, can you share who the camera / analytics vendors are? We could certainly give more pointed feedback knowing what specifically you are evaluating.

March Networks MicroDome with wiretrip analytics was initial scope. However, Mobotix was in here yesterday and demo'd their MxActivitySensor analytic. That would appear to be an analytic algorithm that filters the wheat from the chaff exceptionally well. Additionally they were specific on the effective range - given predictable light levels.

My inclination is to scope an appropriate nighttime LED solution to provide a predictable LUX level over an area that leverages the Mobotix cameras and their MxActivitySensor capabilities.

I am likely to scope edge recording only initially and providing the amplification is sufficient from their cameras, I would use a pre-recorded message stored on the cameras to broadcast a 'get lost' message upon motion alarm.

The presence of permanent nighttime lighting in conjunction with motion detected video initiating audible alarm 'should' be sufficient deterent for transgressions at a very remote gas utility station!?

I think I will lose the photo-electric solution as the sole source for intrusion. I would be lighting at night regardless.

So many choices?!

Brian,

Late breaking news. I owe VideoIQ a public apology. Upon further investigation and troubleshooting in our shop we found the cameras were connected to a faulty client PoE switch and that was causing the harddrive failures.

My bad!!

RJ -

Thanks for investigating/clarifying. Our cameras keep the HDD turned off the majority of the time and are buffering video to an internal SSD first. When that SSD fills up (it's only 2GB, so every few hours), the HDD gets turned on, and data is moved from the SSD to the HDD, and then the HDD is turned back off. When the HDD spins up, there is a temporary increased power demand (negligable for the most part, but there), if your power supply is weak or overloaded and can't supply the extra power, the HDD won't spin up. The camera will retry the operation a couple of times and then give up and mark the HDD as "failed", which sounds like what you saw here.

It's also worth noting that even if the HDD were truly dead, the rest of the system would operate normally, even if rebooted. You would just lose the long-term data retention, but video/analytics/alerts/etc. would continue to function normally unti you were able to repair/replace it.

To get to some of your actual application notes, I'm not a big fan of "tripwire" style analytics. IMO many vendors use this to compensate for lower quality analysis. The logic being that if we only look in a narrow area for objects moving in a particular direction, you'll cut down on the chances that you get inundated with false alarms. The catch is that the camera has to "see" the person specifically while they are crossing the tripwire point. If anything is preventing the analytics or motion analysis software from getting a lock on the object (rain/snow/fog/haze, vegetation, shadowing, etc.) then you won't get an alarm. We tend to prefer settings the entire zone inside the fence as a secure zone, and alerting on any person activity in that zone. In ideal cases, you get an alarm the second the person crosses the zone boundary, much like a tripwiare. In sub-par cases, if the camera doesn't "see" the person until they're already in the zone, you get an alarm a few seconds later, but at least you still get something. In order to do this though, you need to be able to differentiate people from non-people with fairly high accuracy. This classification accuracy is where most other products struggle in outdoor applications.

Just some food for thought.

Note: We are wrapping up testing of Axis' VMD 2.1 and are starting tests on Mobotix's analytics and Axis TripWire. Results to be released over the next few weeks.

John when you test cameras do you test them in the default state? In other words you don't alter the settings before testing? The reason I ask is that our Mobotix sales rep said their cameras are not shipped with setting optimized - unlike Axis for instance (not that i know myself, I was told this) therefore the test results that IPVM get from testing Mobotix are not optimum results because the camera does not come out of the box in an optimum state?

I don't mean to raise anyones hackles (yours or Mobotix), I was just curious.

I think you just hijacked your own thread :)

Lots of manufacturers claim that their cameras need to be optimized, especially when our test results show poor performance. This rose to a boiling point a few years ago so we did an 'optimized' camera shootout. The reality is most manufacturer recommended optimizations did not make much of a difference.

I am not sure if Axis' official position is that they are optimized or not out of the box. Certainly Axis, like Mobotix, has lots of controls to adjust imaging quality.

As for your Mobotix sales rep, what exactly did he recommend to be optimized? In general, I don't trust sales reps as they typically know very little and will say stupid shit to make their competitors look bad. That said, let me know if he offered any specifics. We can also reach out to Mobotix management for a more 'offical' recommendation.

John, This was a random conversation that came up over lunch - just the two of us. I don't recall that he pointed out any particular adjustment. He indicated that IPVM tests cameras in their 'shipped' state. That Axis (as an example) apparently ships their product in a state that is reasonble optimized.

His point being that you don't get an apples to apples test because a default Mobotix camera would be at a disadvantage in such circumstances because their default configuration is hardly optimum.

Our Mobotix rep (in my opinion) is a hyperbole-free kinda guy. He worked for the San Diego PD for many years and got pulled into the video surveillance from the client side.

Understand, I don't have a dog in the fight. However, understanding what is done...or not done in setting up equipment prior to testing is likely to have a significant impact on both the results and how people like myself are influenced by the results.

Must be challenging on your end.

It still begs the question what they want to optimize. What we have found is that most optimizations have significant tradeoffs at other times of the day. So you make it better for a given moment, but 6 hours later, the image quality is a lot worse than if the 'optimization' was not implemented.

Also, for low light, there generally are no 'optimizations' that do not involve lengthening the shutter which has bad side effects of their own.

The one class of optimization I do hear people discuss is brightness/saturation/contrast, but when we have done those, it might be personally preferable to some people but we have not seen it make a difference that a face or license plate could not be caught with the default but could be with that optimization.

We are happy to try out 'optimizations' but manufacturers need to be specific in what steps the optimization consists of. My push back is most of the times vendors throw out 'optimization' in a vague, undefined manner.

So tell him IPVM wants to know specific optimization recommendations.