Swann/Hik and Dahua 1080p license plate clips

[MaxIcon 0]

Ok, here are some license plate comparisons between the Dahua HFW3300C and the Hikvision/Swann 1080p SWNHD-820CAM cams. Let me know if there are problems with any of the uploads, as keeping this stuff straight is a bit of a challenge!

 

Likewise, any feedback on my misassumptions is good. I'm not a license plate capture expert, so any thought from those who have done it with inexpensive cameras would be great.

 

Both are set at 1080p, and the Dahua is zoomed in to 3.3mm (widest angle), which closely matches the Swann factory lens. Both are in auto day/night mode until the forced day mode clips. The cameras are perched atop a 6' ladder, so they're not quite as high as when mounted under the eaves.

 

I used my 2004 Miata, which has non-halogen headlights and a non-illuminated front plate. I pulled in from off camera, pulled into the driveway, and backed out again.

 

 

Since both cameras are on at once, the night mode shots have IR on from both cams, and you can see the effects of the different responses of the 2 cameras to changing light conditions:

 

- The Hik drops into day mode more readily than the Dahua, which switches off its IR. On some of the Dahua shots, you can see the IR brightness on the license plate change when this happens.

 

- The Dahua rarely drops into day mode, but its IR is independent of mode, and is strictly controlled by the photosensor on the IR board, so it'll switch on and off as the headlights get closer and further away. On the Hik shots, you can see the IR switch on as I back out of the driveway, lighting up the plates more (and usually washing them out).

 

 

I took a lot more clips than this, with a variety of WDR settings (for example), but these are the best examples of each mode on each cam. Overall, there are a few general conclusions from these tests regarding using these for night license plate capture:

 

- WDR was useless in night mode. More expensive cameras have better WDR, and it may be better in day mode. When I get time, I'll test this.

 

- HLC on the Dahua was useless in night mode. This should be what HLC is good at.

 

- BLC was marginally useful, but the response times on these inexpensive cameras to dynamic lighting changes is too slow to guarantee useful captures in fast moving situations.

 

- IR did more harm than good, again due to the poor responsiveness.

 

 

So, to use an inexpensive home camera for capturing license plates, you need a few conditions to get any kind of decent results:

 

- A good choke point. The driveway is the best possible, as you can be sure exactly where the cars will be. These camera locations would be useless for cars parking on the street, or pulling in across the street. To get street caps, you'd need cameras pointing up and down the street, zoomed in on the street width. This will depend on your home's layout, of course.

 

- A lens zoomed in on the choke point to get the pixels per foot needed. With these 3.3mm settings, even when the exposure's good, the resolution has dropped enough that a plate at the end of the driveway is not very readable. Zooming in to cover just the driveway would be a big help here, allowing both better exposures and better resolution. The Hik's lens is glued in, so it's not optimal unless you want to hack it. When I do the next round of tests, I'll zoom the Dahua in and test this setup.

 

- Fast enough exposures to avoid motion blur, though a car in the driveway would need to slow down, stop, and back out, so there would likely be some good frames. This would matter more with street captures.

 

- Fast enough response to changing lighting to avoid washout from headlights or IR reflections on the plates. My Vivotek switches in and out of night mode much more quickly than either the Dahua or Hik, so it may give better results, especially if the car's lights are off.

 

- Lighting to ensure the plate's illuminated. The daytime mode gave the best views by a long shot, but if the car's lights were off before they turned into the driveway, the images would likely be too dark. I'll test this in the next round as well. Some cars have plate lights front and back (and some states don't require front plates, of course), some cars have much brighter headlights, and in general, lighting is unpredictable..

 

Buellwinkle posted some suggestions for reliable license plate capture, which include expensive IR and more specialized cameras. I believe a home camera can give decent night time plate capture if the settings are tweaked for the surroundings, but like all night performance on inexpensive cameras, there are trade-offs that more money can take care of.

 

Anyway, on to the clips!

 

Here's the Hik, blc enabled, set for "down":

 

The Dahua, blc enabled, driveway selected for blc:

 

And the Dahua, hlc enabled on medium. The Hik doesn't have an HLC setting:

 

Next, the WDR settings.

 

Here's the Hik, WDR set at 50. Higher WDR made the image noiser, and lower had less effect. It wasn't much use at any setting:

 

And the Dahua, with WDR set at medium. Again, not much use on any setting:

 

Next, the clips that finally look like they're going to be useful! I tried some fixed exposures, as high-speed fixed exposures are used sometimes with powerful IR illumination, but they weren't as good as leaving them in auto exposure.

 

I adjusted the gain so both cameras were about the same when no cars were present. Turning the gain up towards 100 increased the noise quite a bit and made the plates harder to read as a result.

 

Here's the Hik, day mode, auto exposure, gain at 68 (to match the Dahua at 50):

 

And the Dahua, day mode, auto exposure, gain at 50:

 

For the rear plate checks, I set it fixed at 1/30 second, 50 gain, day mode, and backed in to the driveway, to see how it handled the different lighting of the rear end. I'll check this again on auto exposure, but it should be good.

 

Here's the Hik, backing in:

 

And the Dahua, backing in:

 

This is the kind of thing the manufacturers should be posting on their websites, but you can see why they don't from these results.

[Q2U 0]

...This is the kind of thing the manufacturers should be posting on their websites...

 

This is the kind of thing that makes your neighbors worry about you MaxIcon.

 

But seriously, great testing and really good stuff " title="Applause" />

[dexterash 0]

Due to the construction of a LP, no all-day/night, all-around, almost all-auto camera can help.

 

Why? Because, as you could see from your last videos, they need special conditions and need to be "tricked". And this is 1/3 of a solution to correctly and clearly capture LP.

 

I don't consider this a manufacturer problem. This is something general that happens to all cameras.

 

When ANPR is used, usually it involves two cameras: a "general" camera and a dedicated camera that will be used only for LPs. Dedicated in a tricky way, since you can't rely on any auto-algos (awb, wdr, blc, hlc etc).

 

Anyway, a very nice work you put there!

[zikronix 0]

Was BLC enabled in the camera or in BI. I don't recall seeing that setting in the camera

[flynreelow 0]

Great work! Where in California do you live?

 

Here's the Hik, blc enabled, set for "down":

 

 

The first clip works nicely. I was able to capture the full license plate as the IRs turned off. What exactly does blc do, and what other settings are tied to this?

 

Its really to bad WDR doesnt work to well on these swanns/hik cameras. Kind of a shame, as WDR on nice cameras really help out reading license plates at night.

[MaxIcon 0]

Thanks, all, this is something I've been interested in, and it was a good opportunity to run the tests. This is all in Silicon Valley, where we freeze when it gets below 60, and bake if it's above 80.

 

BLC (back light correction) uses a specific area of the image for setting the exposure speed and iris (if the camera has auto-iris). Usually, it's used for things like watching a store's door, where it's brightly lit outside and the entire store front is glass.

 

With normal exposure, it adjusts so that the whole image is exposed correctly, so someone walking in the door will be dark due to the light from outside. Setting BLC around just the door will adjust exposure for that area, which will make a person standing in the door exposed correctly, but everything else will be washed out.

 

That's the main difference between BLC and WDR - BLC only cares about the part you tell it to watch, while WDR looks at the illumination of the entire image and adjusts for better exposure balance across the whole thing. WDR is much better when it's well programmed, but it's a good bit harder to do right than BLC.

 

BLC is enabled in the camera. This is one place where the Dahua is superior to the Hik - the Dahua lets you draw a box around the area you want to use, while the Hik just says up, down, left, right, center, and doesn't show you what area it's using. There are no other settings, like the 0-100 slider on the Hik or the Weak-Medium-Strong on the Dahua for WDR.

 

For the Hik, I just tried different settings until I got the best one, and that was down, not center as I expected. The IR from the Dahua may have contributed to the problems, as seen in the clip you got the license frame from.

 

I have some other tests with the Dahua in a static environment that I'll post up when I get a chance, as they give a pretty good example of the differences in the modes. In those, BLC worked best overall in difficult lighting, as long as you could keep the lighting in the BLC area.

[buellwinkle 0]

I get great license plate capture in pure darkness at about 40' using the Axis Q1604 with VGA resolution with a 5-50mm Fujinon set about 30mm. We use a RayTech RM100-30 illuminator. Very clear crisp plate numbers.

[buellwinkle 0]

Here's a picture of my plate as it's driving by at 20-25mph at 7:53PM last night.

 

[shockwave199 0]

And that can be had for the mere price of....over 2k!

 

Looks good though. What happens at 30mp or so- the image breaks up too much?

[buellwinkle 0]

I get clear images up to about 40mph, after that you get motion blur. The camera is not terribly expensive, I think it's about $800, the lens is about $100, the case can be expensive if you buy Axis like we did but you can buy a case for $50, so under a grand for the camera. The illuminator was expensive, but you may be able to find less expensive illuminators. It actually did pretty well even without the illuminator, maybe because rear plates are typically lit.

 

There's 2 tricks that I've learned -

1. fill the frame with just the bumper, nothing else, meaning get a nice telephoto lens. Resolution does not matter, you can have 10MP, but then you'll be tempted to cover a larger area and that throws off light metering. You can do good LPR with just VGA resolution, so get the camera with best low light performance instead of resolution.

2. have a powerful enough illuminator to compensate for bright headlights/taillights. You can see in my picture, it looks dark, but the illuminator is so bright, it makes the bright brake lights seem like a glow. The exposure is set fast to compensate for the bright illuminator giving us even lighting. Also lets us capture a plate of a car going up to 40mph without motion blur.