License Plate Capture Night Time Challenge

John and fellow forum members, I have many outdoor installations that I would consider quite typical - sureveillance of drive entrances and driving lanes, where the customer is interested in capturing the license plate for potential indentification. Speeds are typically under 10mph. They/I don't require high-speed resolution or LPR analytics - just adequate image capture.

Camera installations are designed to provide sufficient pixel density to adequately resolve these license plates - BUT this only works during the day, with adequate lighting.

The customer most often has even greater interest in vehicles passing these cameras at night, for obvious reasons. I have 'experimented' with various camera setting adjustments, and am now using mostly AXIS Q1604-E cameras for many reasons, including hope of improving the quality of this scenario. I have experimented with IR lighting, but as vehicle headlights include such a large component of IR, it has not proven very helpful so far.

I am sure many others have this customer interest. Any thoughts/pointers on how it can be accomplished with affordable technology - short of $5000+ LPR cameras designed for toll booths?


Michael, have you seen our License Plate capture shootout / test results? You might find it interesting though our results show similar issues as yours.

With the IR lights and the Q1604, have you tried a faster shutter, like moving it to the 1/100s max? This will likely help cut down on the glare from the headlights.

The basic approach for what you're doing is a B/W camera (D1 or ideally megapixel) with an wavelength filter in front of it that ONLY passes IR light wavelengths (the inverse of a typical IR cut filter).

Then, use a strong IR illuminator (Iluminar, RayTec, etc.) near the camera pointed towards where the camera is looking.

During the day, it will work the same, but the images will be in B/W (with the IR wavelength filter in front of the cam you won't ever get a color image), at night the illuminator will kick on to flood the scene with IR. This will help by filtering out the visible light from headlights and taillights, and you'll find that most plates will be very reflective and illuminate nicely. There are some states with uncooperative plates (CT comes to mind), that make this problem a lot harder.

Brian, we tried using wavelength filters to block visible light and were dissapointed. The problem we found was that the headlights generated a lot of IR illumination as well as visible, so the glare/blooming remained.

John -

I've had OK luck with that method testing NH and MA plates. Granted, it takes a little bit of experimentation, and it's not perfect. But in the spirit of the original question, which seemed to be around finding a budget LPC solution, that would be my recommended path of experimentation.

I left out things like shutter speed and the like, assuming (perhaps wrongly) that OP already had those aspects covered. Your tests though are certainly worth reviewing to anyone trying to DIY an LPC solution.

I prefer to read the rear license plate. Less problems with glare and more likely to capture a plate. An informal survey we took in our parking areas yielded between 15% and 20% of the vehicles had no front license plate, despite California law.

Carl, I agree - rear is definitely easier. It's hard to find confirmation about what the laws are for different regions. However, I found this one source that says, "30 states require both front and rear license plates, while the remaining 20 only require rear license plates." However, as you note, if people violate that law, you have an issue.

Thanks for the feedback/confirmation of my results to date.

No, John, I did not see that report, although I did use the 'search' feature on LPR and got many, conventional 'real' LPR articles - maybe I got lazy in reviewing the search results.... I should have known you'd have covered this somewhere.. Great report, as noted many times.

I have 'tinkered' with shutter speeds, but not as thorough as I might - 'real-time' opportunities on these remote installations makes assessing multiple variables a bit challenging. Assuming WDR on?

I have tried the IR pass filter route as well - same results - too much IR in any type of modern headlight.

Tried white LED - helps. I haven't tried - but might - using white LED lighting with IR block filtering (or an even broader higher spectrum filter, if available) to perhaps further improve image quality and lower lighting power/cost requirements - the thought being to reduce the 'impact' of the headlights on on total image lighting. Any thoughts on how that might work?

And given the 'noise' issue at night, would lower resolution cameras work better, assuming FoV is reduced to achieve adequate pixel density? I haven't tried that yet - maybe even a lower cost, non-D/N camera like the M1114-E with white LED lighting?

Obviously, the 'problem' with white LED lighting (besides cost) is loss of stealth - although there are usually limited alternatives to access for vehicles. Mounting it away from the camera might reduce exposing the surviellance installation, but increases installation cost and complexity, as you've mentioned.

John, you mention rear plates working best above, but in rear license plate test scenario in ambient lighting you show poor rear plate results. I am assuming your stating here that IR lighting improves rear plate clarity? Would white LED be of any benefit? My 'test' site is nearly all head-on pass-by shots, often with trailers... I may have to re-think to focus more on rear views, if I can get them.

Michael, my company just recently finished an LPR project for a local apartment complex. They wanted license plates capture for both incoming and outgoing traffic. The hope was to get both the front and rear plates. The rear plate was captured as they arrived. The front plate (if any) would be captured upon departure. Lucky for us, they had one way in, which was a single car width lane. The outbound was also a single lane and the only exit. These lanes were separated by a section of grass and landscaping.

We did quite a bit of testing of different cameras in preparation of this install. What we ended up using was all Nelly Security/Dahua gear. We ended up using a three camera setup for the LPR cams. One 3MP dome for inbound, one 3MP dome for outbound, and a 1.3MP dome for overview. These were all mounted to a 14x10x7 fiberglass box mounted about 8ft off the ground. The pole we attached to was the light pole for the complex. We were able to grab power from the pole itself. We used Ubiquiti radios to transmit the data back to the complex office, which was about 150-200ft away from the pole.

During our testing, we used a RayTec white LED light for illumination, but ended up not needing it for the final install. The light on the pole we mounted provided enough illumination. We had the complex rotate the fixtures about 45 degrees from level towards our monitored area. This gave us about 100 lux of light at the ground where we hoped the cars would pass over.

Our cameras were equipped with varifocal lenses that gave us up to 12mm of zoom. We were able to get our FOV to just the width of the lane, which optimized the amount of pixels per foot the cams would give us. The cams were set to 2MP, or 1080P, which was the max our NVR would record. We ended up setting the shutter speed to max out at 8ms and turned the gain all the way down to 1. These cams also offered WDR, and we set it to the Strong mode. We also forced the cams to color mode, even at night. We found that with the color info, we were able to read the plates better. B&W did not offer more clarity, which we assumed going into the project.

The setting above will give you a very dark image at night, which will not give you make, model, color of the vehicles. It does, however, reduce the headlights / taillights effect to almost non existent. In fact, you can almost see the headlight elements themselves at that fast of a shutter. The reason we found we needed that fast of a shutter speed was due to two factors, the headlights and motion blur. With the 7ms setting, we could read plates up to 45MPH with no sign of blur. That was the limit of how fast we could drive in the test area we worked in. In the installed environment, we couldn't imagine a car getting about 20MPH due to a T in the road with stop signs and speed bumps.

Due to the fact that the complex also needed make, model, and color info of the vehicles coming and going, we added the third overview cam. This 1.3MP cam was also set to color mode, for obvious reasons, but we let the shutter and gain settings remain in auto mode. This camera is able to give us very clear images in color all night long, without much blur at all.

In the end, the setup is working well. We don't have any LPR software running yet to index the plates, but that might be a project the complex may decide they want to invest in later.

Very informative, Gary. I haven't tried Dahua, but I see they have some aggressive pricing. I assume you must be pleased with performance.

Hadn't thought about running gain way down - at the obvious cost of overall image viewing. A very innovative approach.

As you forced to color (I assume internal IR block filter), I assume the integrated IR lighting elements (in thier bullet models, at least) must be of no value. Did you turn it off?

Thanks so much for this info - I'm going to give this a try right away.

Gary, nice feedback and description!

Michael, it's worth testing out what different level of gain control settings will do. The main risk is making the scene very dark (e.g., see our gain control test results). Obviously this depends on how far you go, the relative brightness of different areas of the scene, etc.

There isn't a specific setting to disable the IR, but it wasn't needed. The cameras have an IR cut filter in color mode, so any IR there wouldn't make it to the sensor anyways.

The Camera I like for this is the Bosch reg , reg-x LPR Cameras, but I think its how much you want to spend determines the Type and Quality you use. Lost many projects due to proposing quality over cost .

A few thoughts:

1. The higher you can mount the cameras (within reason, to not get too sharp an angle on the plates), the less headlights will be a factor, as they tend to be aimed forward. I'm thinking of the toll cameras on the bridges here, which are probably a good 10m or more above the road.

2. The more illumination you can throw at the plates, the higher the shutter you can use, the less ambient light/headlights/etc. are a factor. Remember that license plates are USUALLY designed to be highly reflective so they should still show up well, if the camera is at a higher angle.

3. Also, the more illumination you can throw at it, the smaller the aperture you can use, thus increasing DOF and extending the area that will be in focus. Of course, it's a balancing act between aperture and shutter speed, so the more light you can put on the plates, the easier that balance will be to find.

In short, the more of YOUR light there is on the plates, the more control you have under all conditions. If you can exceed (or at least come close to) bright sunlight, you don't have to worry about playing with changing gain and iris and shutter settings throughout the day.

Matt:

1) If you speed up the shutter speed, headlights are no longer an issue at all. However, strong angles will make it tougher to read the plate.

2) Your camera and light source should be as close as possible to the camera. Having the light offset at another angle will change the amount of light that gets reflected back to the camera. The further away the light source is, the more light you will need. You should be careful not to use more light than needed as it might blind the drivers. In fact, in our install, we used existing lighting, just aimed it in the same plane as the cameras.

3) Most people don't have any control over aperature, especially if budget is a concern (as the OP stated), so speeding shutter and controlling light is of utmost importance for keeping your budget low.

Here is a shot of a plate as it passes well after dusk:

Gary: looking at the dark 1/125 image you posted from a EyeSurv branded Dahua HFW3300 2 months ago (sorry for the late digging up), that plate can be read by a human - does a typical LPR solution decode a plate like this into text?

We come across situations all the time where customers look for a software add on to do "optimistic" LPR from existing HD IP cameras deployed across the site without extra hardware/install investment. There are many providers including CitySync and Luxriot that offer pure software solutions. Has anyone deployed them this way (i.e. without deliberate design and positiing of camera for LPR) and obtained reasonable results (80%+ capture rate for example I would consider reasonable). Here we still assume the vehicle will be fairly "head on" to the extent that a human can read the plate (like Gary's image from 2 months ago).

Gary, thanks for sharing that image. What's the shutter speed for the image above? Just curious how fast it is?

1/125s? That makes sense. Thanks.

Btw, for readers wanting to learn more about shutter speeds, see our shutter speed tutorial.

Matt, we tried what you describe above at a college, using IQeye Cameras, spectradome spectra3 cameras

Found in our test the best result was to have as direct of a shot as you could get ( centered approach ) and lots of Bosch illuminators , Slight angle to reflect noise.

This was a while back, not sure what the new management did with what we installed .

Michael, this is a fairly inexpensive solution we have been using that works pretty well for slower vehicle speeds with a narrow FOV (narrow lane)

Messoa SCR505 attached to one of their encoders. Note that we tried this camera with an Axis encoder and their was a compatibility issue so we went back to one from Messoa.

Camera is mounted at about 2' off the ground and is about 1 foot to the right of that parking line you see in the lower left hand corner.


Brian, thanks. Online pricing appears to be ~$500. Is that a fair range?

Also, what was the compatibility issue with the Axis encoder? For a fixed camera, you'd guess issues would be minimal.

That seems about right, maybe 450 through distribution I think.

The problem had something to do with the encoder's method of deinterlacing perhaps? So long ago it's hard to remember. We were getting distorted images of some sort, the last thing we thought it was going to be was the Axis encoder but Messoa said theirs would work so we reluctantly tried one of theirs and it did do the trick with no other modifications to the set up.

Can you make regular cameras work for ALPR solutions? I know you can with the right amount of light and filtering as proven here by some of you. I have been experimenting prior to the last 8 years which is when ALPR cameras became more affordable and readily available.

I do not mean to be critical of anyone here. Lately I am surprised at the level of experimenting and the amount of time people put into a project when it comes to ALPR applications based on the availability of "guaranteed to work" products that are more easily deployed. It would seem that the best camera for any project is always the one with the best light. If there was ever a "guaranteed to work" type system it would be systems engineered for the purpose. Typically, these systems have all the lighting issues worked out for capturing a license plate.

It would seem that for consistency, we would use systems with the physics part all worked out. The intent is to be successful in the deployment of systems so it doesn't cost us the intended benefit during or after the implementation. Is it worth it to continually try to make something work that isn't designed for that application? something that isn't easily repeatable for the next project? License plate capture is physics. Are there are best practices written for ALPR applications?

David, can you share examples of ALPR cameras that are affordable? True ALPR cameras, like Genetec's, are super expensive.

License Plate Capture cameras, with no analytics, no automatic recognition, are less expensive, but typically not cheap. The inexpensive ones are mostly analog.

That said, I don't disagree about the value of using a camera optimized for license plate capture. The amount of time and energy it takes to optimize an COTS IP cameras into an LPC one plus the amount of issues one still faces makes does so typically a losing proposition.

That may be a good research project for IPVM. I recently had a chance to work with Genetec's product. It is thoughtful and seems to work above the average. The Bosch product formerly Extreme CCTV is priced much lower and is lower tech but does the job if properly installed (following best practices). Elsag's add on to Milestone and ONSSI has gained popularity recently. Prices seem to range for a single lane of capture from $700 to $27000 depending on what category the application needs fall into.

There are products specifically designed more specifically for law enforcement integrated to work with online databases like NCIC. Parking is a large market for plate capture with a manufacturer like Federal Signal/PIPS. There are the others that are more security focused like Bosch, Elsay, Genetec, Inex Zamir and PIPS also plays here. Then there are many more designed with transportation applications in mind that are far outside of what we do for security.

Some manufacturers of these products include Accenture, Perceptics, Federal Signal/PIPS Technology, Raytheon, Bosch formerly Extreme CCTV, Mvtrac, Zamir Recognition Technologies, Everview, Kapsch, NDI Technologies, PlateScan, Inex Zamir and Elsag. There are at least twice this many others out there all saying something about how great they work and they certainly may all work but how well?

Recently I was part of a shoot out that involved 11 manufacturers. 1 out of those 11 worked...at all. This may have been for any number of reasons. Sales people showed for the demo and they couldn't make it work, the product didn't work in those lighting conditions, or environmental settings, the product was damaged in transit, the computer CPU failed, the Linux DOM had a static surge causing it to fail, etc. etc. - I think I heard it all that night!

My advice, whether you are a reseller, end user or engineer/consultant, is get a reference to talk to an existing user to see what they have to say, take a field trip to see an existing system, or at least ask the manufacturer to perform a proof of concept at your facility to see it work. Who uses ALPR for security? Airports, universities, home owners associations/gated communities, high net work individuals, cities, parking

David, this discussion is about license plate capture, not ALPR. LPC is when a human reads the plate / number. ALPR is when a machine does so

Of course. My appologies to those I've misled. I meant to be more generic. The component most used in all systems is the camera, a lens, a housing and some type of lighting integrated to work with the camera and lens to enhance the plate capture. Most often they are finely tuned systems. Some manufacturer's like the ones mentioned do offer more than capture. Many offer a stripped down capture only product.

I have no experience in LPC, however if it came about I would research that which has been successful already, I would not try to reinvent the wheel.

You are wasting your time trying to get non-LPC cameras to work in an LPC role 24/7 under all weather unless you implement enough light where it's not much different than a day shot.

Use either an ONVIF compliant or compatible for your VMS, REAL LPC camera, or an analog model attached to an IP encoder.

Don't waste you time trying to shoe-horn a camera into a role it wasn't designed for.

Luis, I have an install with two ordinary 3MP cams doing the job of an LPR cam, 24/7 for the last two months. We have had snow, rain, sleet, etc and no issues reading plates. Light levels are about 40 lux on target. Hardly daylight levels.

40 lux is typical of a brightly lit area like a well lit tennis court or athletic field. For outdoor night time monitoring, that's pretty good lighting. The average parking lot or street is closer to 5 lux - a big drop from what you are facing.

Well, at least enough light.

Well, let's just not confuse 40 lux with daylight (20k lux)...

First, read the Bosch/Extreme patents - lots of good info.

Here is what I found works

1. You need a VERY sharp cut filter. Not just some random night vision material purchased off ebay or a $75 filter for a DSLR camera. I don't think it is so much IR bleed as just needing to have a proper cut slope.

2. Get a good camera with the controls needed, e.g. Axis P1344/1346. Analog cams provide better contrast on the plate. BUT, this was not worth the reduction in resolution. You will need a little more IR to make up the loss in sensitivity when using an IP camera. BTW, Axis encoders suck because they have a deinterlacing problem - i.e. it does not work.

3. Turn down the AGC on the camera somewhat.

Undisclosed, thanks for the input. Very interesting.

The one thing I am not sure how to handle is getting a 'VERY sharp cut filter'. Cut filters are typically built inside of cameras, so it's not something that can be changed. What's the solution there? Also, are you cutting out IR light or visible light. I assume it's visible light? And do you cut it 24/7 or just at night?

I would expect "Undisclosed" is talking about an IR pass (or "visible cut", as the case may be) filter - if you're using IR to illuminate the plates, you can get tighter control over the exposure by blocking everything outside your illuminators' wavelength. The sharper the filter, the less "ambient" light you have to worry about, and your exposure should change very little with the conditions, even between day and night.

Remember that IR cut filters in the first place are only necessary on color cameras - if you're starting with a B&W camera (or ICR locked in night mode), there is no IR cut to worry about.

We have done extensive testing with using "inexpensive" cameras for LPR, and have fielded many of these now. This setup will capture plates up to 65 mph. If the cam can be mounted at plate height, you can easily capture 2 lanes shooting across the lanes at 35 degrees. If the cam has to be higher, then you're limited to 1 lane. Just don't break the 35 degrees max rule.

The Messoa cams mentioned above use an IR pass filter and work very well; we have made our own for much cheaper.

-- Bullet Cam with built-in IR (needs to be long range IR, like you'd find on a 5-50mm cam or so)

-- Take the front off and put in an IR pass filter. If the IR's are 850nm, then you need to use an 830nm pass filter.

-- "Lock" the camera in night b/w mode, either internally or by covering the photo-eye sensor. Important b/c on a cloudy day you'll need the IR.

-- FIXED shutter speed at 1/2000 or 1/4000

-- You can tinker with contrast/brightness/IR levels etc, depending on the cam

Your eyes will pop out at how sharp the plates are. Virtually 100% capture at up to 100', with the right lens/IR combination. You'll need a separate camera for overview.

Wow, Greg, that is a great example and finally someone who puts up.

Do you have a reliable (and reasonably economic) source for the 830nm filter?

Tango, we bought IR filters from Midopt but were not terribly happy with the performance. Even with them, the headlight blinding really did not dissipate significantly. Evidently, headlights put out a lot of IR light as well so it's was still fairly blinding.

Human readability was the goal for this project. The customer had a limited budget that didn't allow for additional software for recognizing and indexing plates. There honestly isn't enough traffic to justify the additional costs involved in an sophisticated system.

Maxmax.com for IR filters.