Triggering for Rear License Plate

[icpilot]

Reference this photo from the IPVM tool...



Capture targets are: 1) northbound traffic after they cross the crosshatched area, and 2) eastbound traffic immediately east of the intersection. Generally, it is within the area as shown for the camera.

Software: Using Blue Iris with Plate Recognizer.

Platform: I9-9900K w/ 32GB RAM, dedicated to BI

Camera: I started with a Frankenstein Hikvision 2085 bullet, modified with various lenses (25mm f2.4 lens at the time of the IPVM screenshot). Daytime success rate was at best 75 % and nighttime capture was abysmal. I now have the OPTION of either the IPC-HFW5241E-Z12E or the IPC-B5442E-Z4E.

There is an overview camera located in the same area as the LPR camera. There is also an illuminator installed at that same spot. The illuminator (Amazon.com : CMVision IR110-114 LED Indoor/Outdoor Long Range 200-300ft IR Illuminator with Free 2A 12VDC Adaptor : Infrared Motion Detectors : Electronics) is advertised as having 200-300 ft range and is pointed toward the intersection. I can clearly see the IR illumination as far away as the sidewalk and into the street, but far short of the intersection.

I live in a state with only rear license plates.

I am finding the rear license plate capture to be particularly challenging. I need the alert image to contain the license plate for Plate Recognizer to work properly. While traffic generally slows for the speed hump (cross-hatched area in the photo), there is little consistency to how much they slow. Some barely slow at all and others will slow to a crawl. Nighttime is terrible due to headlights setting off triggers before the vehicle appears in the image.

Using BI with Plate Recognizer, I think the process is: Motion -> Trigger -> Deepstack -> Confirm -> Alert -> Plate Recognizer

As this particular street is fairly busy, I get numerous "Occupied" results from DeepStack in the triggers each day. Maybe 5 to 10 percent.

For those with Dahua/Empire/Loryta cameras, are you using ONVIF triggers? Are you using IVS settings and if so, which?

Anyone using BI for rear license capture, how did you set the trigger?

Any suggestions welcomed.

 

[wittaj]

One thing I found with the IVS is that it is very good UNLESS it doesn't have enough time to review the object and determine if it is something to trigger on.

In most use cases, it is not a big deal, but for LPR and how tight we zoom in, then it has troubles for some angles.

And night time will be almost zero for LPR and IVS because the image is so dark so the camera does not see the outline of the car. DeepStack will suffer as well because the image is black so it won't recognize a vehicle just by the head or tailights.

So you would have to run simple motion detection for the camera in this scenario, but even then that can be tricky, so you need to run Blue Iris motion detection instead of camera IVS.

I found this to be the setup that worked best for me. This one zone approach at this location allowed me to catch the front plate of the cars going right to left and the back plate of the cars going left to right. Occasionally I would get a large truck or a very fast vehicle that was missed, but I also run this continuous motion in the event that happens or it missed the motion. And then at night I simply make the object size box a lot smaller like the size of the plate. As you can see from the alert clips to the left that it captures all the cars in the middle of the frame. My nighttime is the same as well. If you are running a 1/2000 shutter then it should take out a lot of the headlight bounce that would come from the street and cause it to trigger early.

If you are only interested in the back plate, then make another zone that is to the left of this and tell it to only trigger for motion from the zones going left to right.

Post your settings for night and the zones you have and we can help you out.

But it does take some trial and error to get this down.

 

[icpilot]

First, the camera. I decided to try to 2MP camera, IPC-HFW5241E-Z12E . Present zoom is probably 2/3 of its full zoom.

Starting simply, so the attached shows Zones A & B, with trigger set to A>B.



Results in the following captures. None of them are reliably capturing the LP with the first Alert image, so I am not yet sending to Plate Recognizer.


View attachment MPVFLPR.20210814_175021_3.mp4
View attachment MPVFLPR.20210814_175021_2.mp4
View attachment MPVFLPR.20210814_175021_1.mp4



































Exposure settings are these:




Thanks for the assist.

 

[sebastiantombs]

Increase the pre-trigger time.

 

[wittaj]

That is the same camera most of us use for LPR, so it is simply a matter of dialing it in to your field of view.

Do not be afraid to zoom in more.

Yep, increase pre-trigger.

Here are my settings, except I am running 1/2000 right now. Change shutter to manual so you can set the gain. That is what is probably getting you at night.

 

[Flintstone61]

is there any possiblity of reducing the angle of the camera, ( mounting location/viewing area) so that the car stays in frame longer to get more useable frames?
I think my z12 benefits from a having like 15% less angle than your graphic.
I also am not challenged with higher speed. it's a driveway to a parking lot.

 

[wittaj]

This is almost the same angle that I helped @ProTapper set up with TeamViewer - wanna fill him in on more zoom LOL.

 

[wittaj]

I would also suggest a curved pattern. With your setup you will get every car coming off that intersecting street and sending images to Plate recognizer that the plate won't be seen.

 

[icpilot]

That is the same camera most of us use for LPR, so it is simply a matter of dialing it in to your field of view.

Do not be afraid to zoom in more.

Yep, increase pre-trigger.

Here are my settings, except I am running 1/2000 right now. Change shutter to manual so you can set the gain. That is what is probably getting you at night.

I think you meant to include a pic, but I only see a filename.

I did increase the pre-trigger in BI to 0.2 sec. I changed the shutter to manual and left the shutter at 1/2000. What gain are you using?

Right now I'm getting early triggers due to headlights, so I may turn up the HLC amount.

 

[wittaj]

Weird let's try again.

My older Z12 gain is 30-30, but the newer Z12 is 80-80. The newer Z12s use a different chipset than the previous ones and the settings on my recent one I got are completely different from my older 5231-Z12 or neighbors 5241-Z12 settings (that are using my identical 5231 settings). So the baseline settings you may see referenced on the site are different with the newest batch.

That fast shutter speed should knock out headlights unless you have lots of cars with low fog lights - those are the only ones that trigger the picture early.

Post the motion settings page from BI

 

[Flintstone61]

.png file name no image

 

[icpilot]

Weird let's try again.

My older Z12 gain is 30-30, but the newer Z12 is 80-80. The newer Z12s use a different chipset than the previous ones and the settings on my recent one I got are completely different from my older 5231-Z12 or neighbors 5241-Z12 settings (that are using my identical 5231 settings). So the baseline settings you may see referenced on the site are different with the newest batch.

That fast shutter speed should knock out headlights unless you have lots of cars with low fog lights - those are the only ones that trigger the picture early.

Post the motion settings page from BI

Set the gain to 80 at Night. HLC at 50.

BI Motion page here ....



Produced this capture:
View attachment MPVFLPR.20210814_204617_2.mp4



















PS.... Still no photo in your post, only filename. Weird.

 

[icpilot]

And another early trigger due to headlights ....
View attachment MPVFLPR.20210814_204617_3.mp4

 

[wittaj]

Wow that isn't very good. Wonder if the chipset has changed again.

Add some make time, try 0.3. I think that speed bump is going to be problematic for you as that is causing an early headlight bounce off the pavement.

Make the min object size about the size of the plate. It won't knock out the bounce, but a little larger and the make time may get you closer.

You may need to make a custom Deepstack model that you draw boxes around the plates for your field of view so that it isn't sending tons of pictures to Plate Recognizer.

Post a screenshot of all of your camera gui settings.

We may need to drop gain down or iris or increase difference between brightness and contrast.

 

[wittaj]

You could try playing around with the settings of not having BI trigger for the first instance (headlight bounce) and then trigger on the 2nd instance. Depending on how many cars come thru, it may work.

 

[Flintstone61]

I'm not a Deepstack user. I do however have my z12 set to save .jpegs at 0.3 sec intervals at 100%. Would DS be able to use those images somehow?

 

[icpilot]

I'm not a Deepstack user. I do however have my z12 set to save .jpegs at 0.3 sec intervals at 100%. Would DS be able to use those images somehow?

Right now I am bypassing DS entirely. Just using the native BI trigger. Once I get the triggering reliably set, I hope to use BI's connection to Plate Recognizer which depends on a single Alert image.

 

[wittaj]

I'm not a Deepstack user. I do however have my z12 set to save .jpegs at 0.3 sec intervals at 100%. Would DS be able to use those images somehow?

The issue is with how dark the image is at night, DeepStack may or may not recognize it is a car, and if it does, it could identify it as a car early. Just like in the daytime, DS for me can trigger with a quarter of the car. If you are trying to get the back plate, DS blew its load identifying it as a car early.

And then the issue with Plate Recognizer is the one alert image sent from BI per trigger event, so if the plate isn't in the image, it sends it and it counts against your monthly total.

So it is a lot of trial and error to get the zones drawn up properly to time the alert image to have the plate in it.

One would have to create a custom model that looks just for the plates and then triggers the alert image on that confirmed instance. Someone here made a custom model for his field of view that works very well. He said it takes about 3 hours to create it.

 

[Flintstone61]

ok. your last video looks like your getting closer. I could at least see a plate in a freeze frame but the digits were a little muddy.

 

[wittaj]

Yeah the last video is closer - looks like a combo of dropping gain (maybe try 60-60) and lower the iris and increase the difference between the brightness and contrast will pop it off the page.

Try running NR lower and turn off Smart IR and do manual IR and blast them at 100.