IR - detect a (warm) car at 10 metres in a 30-degree cone

[SL4P]

A challenge I'm contemplating.

Outdoor, IP67 permanent installation. 
0-2m (0-6ft) above ground level. Day or night.

How to identify if one or more (motor vehicles) are within the field of view.
Maximum range 10 metres (33 feet)
Engine may be idling or stopped, but active parts (engine, exhaust etc) will be 'warm'.

Some false positives are expected, but goal is >90% acccuracy.
Any thoughts?

[BradC]

Some false positives are expected, but goal is >90% acccuracy.
Any thoughts?

Right up my alley, in fact I recently spent 8 weeks trialing a number of technologies at a large facility to do a similar task. More information on your objective would probably help.
Mine was a bit more complex in that I had to tell a vehicle from a human/dog/plastic bag/cloud/reflection/body of moving water, and I had to determine direction with multiple targets simultaneously. Your task might be a bit easier, but probably not actually that much.

If you are simply detecting a stationary heat emitting object, then it's a bit easier again.

To be honest, at 10M you are probably better using something other than IR, or a combination of technology. Heck a set of induction loops in the road would be cheaper and probably more reliable in the long run.

[miguelvp]

On the cheap:
IR thermometer with data logger?
But finding a 2:1 spot ratio one (for the required 30 degree cone) might need extra optics.

More expensive but probably not prohibitive:
A thermal camera, and that probably is closer to the 30 degree cone required.

And of course software to determine if a car is there or not. For the thermal camera the software will need more supporting hardware and a more complex but more accurate program.

Or do what they do to detect cars on intersections for traffic lights, not IR based as far as I know.


Edit:
Hack this thing, it has the 2:1 spot ratio (hopefully it means 1 meter at 2 meters distance not the other way around)
http://www.ebay.com/itm/DT8260-Temperature-50-to-260C-IR-Infrared-Thermometer-Intelligent-Measure-Meter-/141547588794

[Zero999]

Does it have to be IR? What about radar?

[mikeselectricstuff]

What's wrong with the way everyone else does it - inductive loop in the ground?

[SL4P]

A few good points raised - thanks...
It may be a combination of visible and thermal...

To elaborate on some details that may help.

• The background will generally be ambiguous (buildings and open space / vegetation / fences etc) - rarely other vehicles within about 5-10m. (a POSITIVE)
• I have no direct/immediate access to the property where vehicles may be parked (a NEGATIVE) - so no loops etc
• The devices would in fact be placed without knowledge of the vehicle owners, probably solar charged, and communicate via a low-power mesh network.

Edit: Electric vehicles: Hmmm thanks - not a problem for a couple of years, but a real challenge 5 years from now in this particular situation.  May have to detect moving electric fields!!(?)

I should add this is NOT a motion detection problem - but the presence of a vehicle or 'not' in the monitored location.

[nfmax]

Maybe a circularly-polarized microwave radar? Reflection from metal objects reverse the direction of circular polarization; scattering from dielectrics doesn't (more or less, anyway).

[SL4P]

Hmmm thanks again - Microwave + PIR comnined may be interesting. (No sun problem).

The car be damned - but it makes me think also that oe way of lowering power would be to test at an interval - e.g. every 10 secs, rather than continuously.  averaging that will also help reduce spurious positive reflections.

The cat lives to pee another day!

[G0HZU]

I would have considered using a day/night CCTV camera and then capture and memorise the scene without the car. i.e. presumably there will be permanent background shapes/objects that the car will mask when it is parked and this can trigger the alarm.

So you could detect for the lack of these shapes when the car arrives using some image processing software on a netbook PC that breaks the scene up into a grid pattern and learns what the scene looks like without the car.

To overcome scenery changes due to weather the system could slowly capture/relearn the scene during the day or night. i.e. design a system that accurately detects for the car 'not' being there rather than the other way around.

You could capture an image and log the make/model/plate of the car and time of arrival/departure as well. You could get it to auto email you an image every time there is an alarm (up to a sensible limit )

[SL4P]

An interesting idea - and I should have thought of it myself...
I have a background in video systems!

Small CMOS cameras are cheap enough - we're talking about potentially hundreds of these little gizmos deployed all over the place. - but no pre-defined order or reason in their placement. 

Would still need PIR to identify a 'warm' object so that incidental bodies or parked (cold) cars are not detected....

Keep thinking - many thanks to all contributors

[eneuro]

Some false positives are expected, but goal is >90% acccuracy.

There are many different vehicles-what about bloody flying drone? 
Is it vehicle or not?
Are you looking for detection of something but what about its size?
Bigger car means a lot of metal, so probably metal detection by means of induction might work, but still endless posibilities to confuse such system while there are endless possible vehicle shapes and types, so it is hopless to detect something not well defined.

[CatalinaWOW]

 A crisp problem statement will help others help you, and maybe help you too.   The needs for an automated parking control system are very different from a security related system, or one that is intended to trigger human investigative action.

 You suggest that a video setup that detects a car that has moved into your field is not satisfactory.  Does this indicate that you don't care if someone who has only driven a few dozen meters parks in your monitored area?  Does the setup have to be a singular sensor?  Perhaps if you really care about heat your could have a narrow field of view sensor heat sensor (the thermometer mentioned above) on an az-el mount that is pointed at candidate locations by your wide field of view visible or NIR sensor.   A dual band approach like this could have a much lower false alarm rate than most single band approaches.   Your hint that you don't need constant surveillance hints at many other options.  If you only need to check once a minute or so scanned narrow FOV systems become very practical.  A question thus far unanswered that will affect many approaches to the problem is whether you can guarantee a null condition at set up and/or periodically during operation.

I suspect you don't want to reveal too much of your problem for a variety of reasons, but without such description you will get a lot of red herrings.

[SL4P]

Agreed about having to field red herrings - but I can't reveal the exact challenge - as it is not illegal, but would be misunderstood by the parties being surveilled...

Inductive sensing is out - as already mentioned, because the sensors can't consistently be in a position near enough to the target object to be sure of a field. 
Some may be viable, but >90% no chance, plus civil works to dig-in a loop would not be an option.

Visual-only is out - because some detection may take place in near blackout conditions...  vehicle arrives, parks, engine off, but still in position to be detected as a warm mass.
PIR can assist with this.  Perhaps not full field PIR, but a defined matrix of points with  credibility score to return a pass/fail.
 
The vehicles - for clarification are typically regular passenger vehicles or small commercial vans.
Parked in reserved spaces, at the side of the road, or in other outdoor places.

Sensor assemblies must be as small as possible for covert placement on *any* axis wrt the sensor unit between 2-10m line-of-sight from the actual target vehicle.   
Goal is autonomous operation on solar charged cell pack, with 'simple' low-power mesh comms of ID & status/score data as needed.

Currently this is a concept paper being developed, to be followed by some R&D and a POC if it seems viable at the target price point.  All your ideas are appreciated - so please don't think it's a waste of effort.  To me it's a really interesting challenge to solve a real-world problem that is not in a perfect / controlled environment.  In fact if it looks do-able, it could easily become a crowd-funded project.

[bookaboo]

I'm working on a vehicle detection system, luckily inductive loops are an option for us. I do know that the competition have a magnetometer based product for the same purpose, they have a sensor 20mm diameter x 70mm long, its a home made potted type. We did rip it apart but they had scraped numbers off the chips.

[CatalinaWOW]

Another option would be range finding sonar sensors, which are fairly cheap, small and fairly low power.  Look into Maxbotix (www.maxbotix.com) for examples.  In any sensing band you will need to do delta reading to sense the arrival/departure of your targets.