Video Analytics, what is available to us regular consumers?

[drvnbysound]

Doing a few quick Google searches, I was able to find the types of images I was referring to above.. This is an example of the actual thermal spectrum view:

... and this is a Thermal IR view:

I am not certain on the first image, but the site I pulled the Thermal IR view from stated that image was taken in "pitch dark" conditions. I personally, am used to working with video analytics and the latter types of images. With this setup, I have RARELY seen false alarms based on wind induced movement of tree branches, clouds, etc. With the analytics I have used and these cameras, it almost always takes a person or vehicle to trigger tripwire or area of interest types of rules.

I have worked with models from Pelco, FLIR, and a couple of other companies. Regardless of the specific type, I do believe that any thermal camera would likely be out of the budget for anyone DIY-ing this to protect their home, as I believe all of them are in the $5k+ range. I dont recall specific models that I have worked with before, but here is a link to a Pelco Fixed Thermal IR model: http://www.google.com/products/catalog?hl=en&safe=off&biw=1728&bih=797&q=pelco+thermal+ir+camera&um=1&ie=UTF-8&cid=12514894877034629456&ei=I7E3TYIKwfnwBr2v7PIK&sa=X&oi=product_catalog_result&ct=result&resnum=10&ved=0CHsQ8wIwCQ#

 

[drozwood90]

I too have worked with thermal cameras, so I understand the technology. However, there are different types of thermal cameras, as far as what the output video looks like (either showing a real-time thermal spectral image and others that are filtered such as infrared). Yes, they do "see" branches, leaves, etc, but everything that makes up the "background" of an image generally has a temperature that is not rapidly changing. Having said that, a person walking across the view should show up as a hotspot for the camera, as is generally easily detected with analytics. I will see if I can post some video links as well.

Sounds like you had a poor camera or the AGC was not set to react to the scene properly. There are at least two companies that I can name that do what you are asserting (more of a static AGC). I can't mention names (NDA in place for at least another 6 mo.). --Keep in mind, I wasn't working with them, I designed them. From the component up. Designed boards, control circuitry, power supply boards, etc. The only part I did not have some influence over the design was the sensor head and the optics.

My cameras could see better than 0.01degK differences, so after scaling (since you worked in the field, you understand that you need to scale to the greyscale, or color pallet you are trying to hit) it had "more" information than a visible spectrum camera. I agree, IF you have proper AGC running, you will get hot spots of "things", but keep in mind, there are two times a day (in warm weather) that thermal cameras go blind and require a better ... how can I say this without violating NDA's...viewing of the video stream to properly see. So, what I'm driving at, thermal cameras can be just as blind to stuff, depending on the environment.

Also, tossing this out there, if you are in a desert vs. being in Alaska. Or for visible, being in Alaska and having the analytic software look for anything that is not white.

I also disagree with your assertion that the two type of thermal cameras are thermal and "filtered" IR. The second you mention is near IR. It is NOT IR. There is a reason they require a light source of some kind ("IR lighting"). They are only picking up near IR spectrum (which tends to proliferate the dark better due to it's longer wavelength...similar to how a bass speaker can proliferate walls better than higher frequency tweeters).*** Hence why there is a cut filter on cameras that are like that, or tend to claim "great night vision", or can see in 0.000000001 lux. They are looking in the near-IR spectrum. NOT IR. It's physics. Something that sees Visible can NOT see thermal. The wavelengths are too far apart for CCD type material, at this time. They use the filter to keep the CCD from saturating when there is "real" light (i.e. Sun).

There was a company called Red Lion or something like that (I can't find a link)...at one trade show, they claimed to make a device that you could put in front of a CCD, and it would transpose thermal to near-ir. I never saw it work. Never heard anything about it after that one trade show. Then I left that field. I don't know if they are still out there.
There two kinds of true thermal cameras are:
One requires a chopping wheel of sorts (as it can only see differences - AC coupled if you will **) and one that "sees" thermal (staring style, or DC coupled*). I have experience with both, but designed the second. The first will have all sorts of noise and crap on the screen.

--Dan

* kind of like having an array of thermistors...then you read them all and put them together to "see" a mapping of the picture...like a bitmap
** kind of like taking a "security" sensor, and putting a chopping wheel in front of it. If you do that, you can see the differences between the unchopped and chopped portions of the chopping wheel. When you then take that and map it into a picture, you again see a "bitmap" of what the sensor head saw.
*** this is also why I was able to borrow a camera from work and locate a bees nest in the walls of my last house. The thermal energy proliferated THROUGH the wall of the house and I was able to pin-point the location of the nest. I agree with your assertion, it was easier at night, but it's not that the nest was not visible, but the math for scaling the analog signals from the IR sensor head with AGC, etc. are easier with larger differences.

 

[drozwood90]

Doing a few quick Google searches, I was able to find the types of images I was referring to above.. This is an example of the actual thermal spectrum view:

... and this is a Thermal IR view:

I am not certain on the first image, but the site I pulled the Thermal IR view from stated that image was taken in "pitch dark" conditions. I personally, am used to working with video analytics and the latter types of images. With this setup, I have RARELY seen false alarms based on wind induced movement of tree branches, clouds, etc. With the analytics I have used and these cameras, it almost always takes a person or vehicle to trigger tripwire or area of interest types of rules.

I have worked with models from Pelco, FLIR, and a couple of other companies. Regardless of the specific type, I do believe that any thermal camera would likely be out of the budget for anyone DIY-ing this to protect their home, as I believe all of them are in the $5k+ range. I dont recall specific models that I have worked with before, but here is a link to a Pelco Fixed Thermal IR model: http://www.google.com/products/catalog?hl=en&safe=off&biw=1728&bih=797&q=pelco+thermal+ir+camera&um=1&ie=UTF-8&cid=12514894877034629456&ei=I7E3TYIKwfnwBr2v7PIK&sa=X&oi=product_catalog_result&ct=result&resnum=10&ved=0CHsQ8wIwCQ#

Yeah, your first image proves my point. Poor AGC control. So, they wash out the "background" based on a heat source. Which means a car parked, but stationary can force the camera to go blind if the sun hits it just right, then the car becomes the "hot" spot in the video and you go blind to "people" (car gets to say 140deg, people are 98.6...the people can become background "noise").

The second image didn't load.

However, to your point here, I agree, true thermal cameras are out of the price range of DIY. However, the range of the cameras I worked with are $22-104K+ (which could see the temperature variations of the human eye, down to the pupil from 22KM away). So, could be why my experience with them is so much different.

I will also state, the image you posted is NOT how the camera will look. That's a marketing photo. I've NEVER seen a Pelco or a FLIR at the level you are talking about that made a "quality" image. They are very poor quality and will DEFIANTLY produce results you are talking about, assuming you are not saturating the camera's view. The higher level FLIR (one of my competitors) and the cameras I used to design, they will produce those images you see, UN-shopped. I should also state, the Pelco cameras, they are re-branded FLIR.

Funny thing, my system was so good, that the producer of the sensor head - HAD cherry picked a sensor head and had one of their customers design their demo camera. My system with a NON-cherry picked sensor head was so much higher in quality, the sensor head producer actually asked us to make one for them. We "sold" them one of our off the line cameras. It was still 100fold better than what they were using to demo the hardware.

Just as in statistics, it's HOW you look at the data. Quality of the data means a lot, but so does processing it. That's I guess the WHOLE point I am making. Thermals CAN be better, however, with the "level" you are talking about, I agree...they "can" be better due to how well they "ignore" the background stuff (poor agc). I didn't mean to make a big stink over this, just that what you said only applies to the products at that level, not of the technology (which is FAR superior to visible).

--Dan

 

[johnny]

i have a logitech alert exterior camera and Ive only played around with it inside the house (to watch for mice in the basement) but i like the image quality and integration of hardware and software. I find it loads better than a microseven sony had ccd IP camera that I purchased with not too much research.

since logitech's motion detection software isnt very good at all, Im thinking of using vitamin D or blue iris. i liked blue iris's web server and video streaming, but liked vitaminDs motion detection even better.
One of my ideas is to run motion detection using vitaminD on the low quality stream from the logitech alert, and record the high quality stream continuously on a several day rotation cycle using blue iris.

I also have an external motion detector thats tree, shadow, snow, and wind immune but it wasnt cheap. it is a Optex HX40-RAM thats wireless and battery powered. i have it installed in front of the garage door to monitor for people and cars in the driveway. It's on a nonalarm zone on my Elk but I get voice notification 24/7.

somehow i would like to tie vitaminD, blue iris with the logitech alert, and the motion detector to have complete reliable video surveillance but I'm procrastinating with the camera install while its subzero outside.

 

[electron]

Vitamin D works well with the Logitec Alert cameras, but the Alert cameras themselves can be problematic (especially the outdoor model).

 

[drvnbysound]

Something else to possibly consider..

What kind of pricing do these consumer grade analytics run? Are they open source or have any sort of API? The reason I ask, is simply that if there is any way to integrate analytics from multiple vendors to a single camera, you could potentially have rules that are more effective from a particular vendor. That way you could set various rules on the same camera running different analytics, which could potentially be more effective. Another option would be to duplicate the rules, and have the alerts compared... if both analytics alert within x time (e.g. 1 second) then you get an alert that an event occurred.

Generally, we do the same thing with various sensors on our systems. For example, doing comparisons with video, acoustic, seismic, sensors in a particular area confirm that there is an event occurring, and is probably a high priority. If only one of them is triggered, there may be something there, and it needs investigation, but its not as high of a priority. There are a ton of different ways of collecting, comparing, and presenting data like this, it's all just dependent on your particular situation and how you want it handled (and if you can do these, generally based on the open'ness of said software). It seems that while a lot of software is being written to be integrated into larger systems, a lot is also becoming more proprietary.

 

[electron]

Looks like they just released a new build of Vitamin D: 1.4.2, which improves memory usage and adds a few other features.

 

[Quixote_1]

You guys will probably be interested in this:
http://info.ee.surrey.ac.uk/Personal/Z.Kalal/tld.html

It's an open-source image processing algorithm for object recognition. It can be used with faces as well.