Can a VideoIQ unit differentiate between deer and human?

[DAWO 0]

Hi,

 

We are in the process of setting up a CCTV system combined with perimeter detection to turn on voice messages and light. However, as we occasionally have a couple of deer passing by in our garden, we are trying to figure out how we should set up the system so that it will differentiate between human activity and a deer passing by, which should NOT trigger any voice messages or turn on the lights or wake us, or the neighbours!!!

 

But when I stumbled over a footage on YouTube showing that the VideoIQ system will not be triggered if a dog comes by, I thought that maybe this means that VideoIQ could handle the detection of human vs deer, meaning I will not have to invest at all in any more advanced motion detectors or such arrangements to sort deer from human...

 

But as promises in the past about the capability of motion analysis within the video sector have been over optimistic, I wanted first to listen to you guys in this forum. Maybe someone here has already invested in a VideoIQ system and have the same conditions as we have with deer, cats and dogs occasionally passing by???

 

Any real world experience on this matte would be most appreciated!

 

Regards,

 

Dag

[stallion 0]

Hi Dawo,

 

I don't know the VideoIQ software, but I have worked with several motion detection analytic's. Many of them allow you to define the size of the object you want to be concerned about triggering the alarm/event.

 

Simple ones simply let you draw a rectangle around the space you want covered, and maybe adjust the sensitivity.

 

More sophisticated ones let you draw a coverage area, and draw a box for the smallest object you want to trigger the alarm.. and then the largest object you want to trigger an alarm.

 

A simple analytic that you set up to alarm on something the size of a person will be triggered by a deer.

A sophisticated one with littlest/biggest threshold might not trigger for the deer. You would probably need to test it quite a bit with something that size to fine tune it.

 

It may not be a big deal if it only happens once in awhile. You are likely to get false triggers on occasion with almost any system.

 

Not sure if it helps or not, but that's the basics from my experience.

 

Stallion

[DAWO 0]

Thanks a lot! That helps!

I think I can live with one or two false alarms after a period of "trimming" the system. As the cameras we are concidering are of fixed dome type, I guess it would be not too difficult to find threshold levels for "our" deer so that they do not trigger any events. We will only have three cameras around the house and we are looking for something at the same level as a Sony SNC-DH180, Panasonic WV-NW502, etc. Given these types of cameras and their resolution, which software would you suggest that "best" will help us sort out deer from humans

 

Thanks!

[thewireguys 3]

Yes it can. The analytics are automatic and no setting min or max for object size. Just mount the camera aim and focus.

 

FbAWYB24Qdw

[DAWO 0]

Thanks!

 

That's amazing!! And you mean that it will even work with deer which are a bit larger than those dogs??

[DAWO 0]

So does this mean that the VideoIQ unit is much better when it comes to filter out animal (regardless of size) than a more traditional solution with a HD CCTV camera and some kind of Intelligent Video Management Software???

[thewireguys 3]

Here are some other videos you can watch and see for your self.

 

http://www.videoiq.com/videos.php

[cbodbyl 0]

I haven't used VideoIQ before but I've installed dozens of IOImage analytics encoders(now owned by DVTel). My experience has been that they take a lot of fine tuning to distinguish between animals and humans, though it is very possible. You will need to have a very good understanding of how to configure the analytics. From what I know of VideoIQ they are supposed to be very easy to install with no configuration necessarily. Not sure if that's true or if it's just good marketing. @thewireguys have you installed any VideoIQ cameras? and do they work as well as they claim? I've been looking at switching to them for a while now but I can't justify the expense if they're not going to work as advertised. Would love to hear from someone with some hands on VideoIQ experience.

 

Colin B

http://www.zeecure.com

[thewireguys 3]

Yes I have multiple VideoIQ installs and demo cameras. I analytics work very well MUCH better then IOimage cameras.

[NotoriousBRK 0]

Yes, our system should be able to provide the ability to distinguish between a person and a deer (or other animal) in most cases.

 

VideoIQ does not use "motion detection" analytics, and it does not require you to spend all kinds of time tweaking and tuning the system in place.

 

Basically, we use a "pattern matching" system, where the system learns the background of the scene and compares the speed, shape, size, color, texture map, movement direction/style, and other parameters of moving objects to an on-board database of over 200,000 visual patterns and descriptions that have been learned into its algorithms. Based on this it is able to auto-calibrate to the specific scene/installation where it is installed.

 

ANY video analytics system is going to be dependent on the overall visual quality of the scene, meaning that things like good focus, lighting and lack of distortion in the image will contribute directly to the quality of the decisions it makes. Or, to put it another way, coupling a crappy camera to an advanced system will tend to yield crappy output

 

You can take our encoder and attach it to a good quality analog camera, or you can use our iCVR-HD megapixel camera and get really good images. The HD camera is the most popular choice, and we're currently the only company doing real-time megapixel analytics.

 

To address your specific question though, there MAY be times where it is difficult for the system to distinguish between a human and a deer. A deer walking directly towards the camera is going to have a lot of the same visual characteristics as a human. It is going to be about the same height and aspect ratio, have similar movements, etc.

 

There are things you can do with our system to manage this to a large degree. Rules have a detection sensitivity setting, which allows you to specify how good of a match an object needs to be in order to trigger the rule. You can also set parameters like a detection time. Basically a loitering alarm meaning that it needs to detect the object consistently as a person for say more than 3 or 8 seconds. A valid human will be seen as a human from all angles. A deer only from very select angles. Upping the sample time reduces the chances of a deer triggering a false alarm.

 

Unlike some other vendors, I am *never* going to claim that you can ever get to 100% accuracy in all cases because that is just simply impossible. Hell, it's hard enough to claim 100% reliability from a simple burglar alarm door contact or motion sensor. To claim it for any advanced sensor system like this is just dishonest, IMO.

 

My personal opinion is that our system will provide the best possible results for visible or thermal cameras.

 

BTW, the car wash video clip is from one of the VideoIQ cameras at my house, and came straight off the system, there was no trickery or editing there.

 

Here is another clip of me and one of the dogs at the office:

 

The yellow box you see briefly on the dog is what we call a "suspicious" or "unclassified" object. The system sees that the dog is starting to match some criteria, but not enough for a full classification.

[DAWO 0]

The VideoIQ camera looks really very promising indeed!

 

Is there any way that I can do to make the system even more accurate when it comes to differentiate deer from human once I have installed the camera? Like send in samples of footage with deer and human present in our set-up to VideoIQ?