looney2ns
IPCT Contributor
Review: IDahua DH-ITC-437-PW6M-IZ-GN / LPR-437 - ALPR/ANPR - Traffic & Access Control Camera
- Thread starter Wildcat_1
- Start date
Will be replying this week @Timokreon and will look to assist. Apologies been switching out infrastructure layers which have kept me busy for a few days since postingDialed my 437 in better. Still room for improvement, but getting better.
Timokreon
Getting comfortable
Apologies not needed at all. I'm very appreciative that you help us on this forum at all. I look forward to your nighttime review and will be using any tricks/tips on my setup.
Plus, it's kind of fun, yet frustrating at the same time, trying to get this camera dialed in completely.
It really does seem that it's very finicky about having the proper settings used or the picture ends up a disaster. lol..
Timokreon
Getting comfortable
Another, "maybe" bug. The focus slider reading on the Wizard and the anpr setup button don't seem to match up. I've attempted to refresh multiple times, etc..
Perhaps I'll do a default and start over this week.
Perhaps I'll do a default and start over this week.
Smilingreen
Known around here
Very professional video and instruction. Too bad I live in the middle of no where, 1/3 of a mile off the road, behind a big hill and in the woods. I don't really have any use for a camera like that to play with. That's the price you pay to be able to pee off your front porch and no one to complain about it.
After having the LPR-437 for a month, I finally got it installed yesterday. The camera is in a painted box at the corner of a brand new fence. The aiming is very crude right now - instead of using the supplied mount it is bolted to a 3 x 3 x 8 inch piece of wood with plastic shims to adjust the angle up or down. I am waiting for a Dahua DH-PFA162 from Andy's Amazon site for a final installation. The camera is about 8 feet up and 32 feet from the center of a 36 foot wide residential street. The center of the FOV is approximately opposite the fire hydrant in the picture below. License plate sizes are around 200 pixels across for US plates. Close to an ideal setup.
Box with side removed:
View before mounting box at corner:
The camera adjustment have been to set the video to H.264 and CBR = 10240; set the shutter to manual, 0-1 mS; enable FTP for image storage off camera; add it to Blue Iris; and set the capture area to a diagonal stripe. No optimization has been done.
Results have been excellent so far. No big problems even at night. All of the following images were sent by the camera to a computer, either via ftp or by the browser automatically downloading it.
Dirty plate:
Seems to have problems with speciality plates - tries to read the small letters.. Also has problems with symbols that are on some vanity plates - they are skipped.
Another interesting misread case:
Biggest problem I have is some plates are truncated at the edge of the FOV.
This is despite moving the capture area in from the edges.
I have done nothing except set the shutter speed. No optimization at all so far. Here's a night shot:
I believe this is a paper plate - I have never been able to capture one at night before.
Another paper plate, late in the evening. I do wish the camera would select better frames.
A Tesla with no front plate:
A few comments:
0) The camera should do a better job of selecting which frame to analyze.
1) Most of the web interface seems to work with any browser but the search function asks for old versions of Firefox or Chrome so that an extension can be loaded. Really??? That should have vanished years ago.
2) Under Settings --> ANPR --> LED Screen - what is this?
3) Under Setting --> Camera --> Image --> General - what does the default environment setting do?
4a) When looking at the camera from a browser, it started downloading the files to the computer using the storage path in 4b)..
4b) Under Setting --> Storage --> Storage Path - why is there no way to use the %09 macro to include the license plate in the filename? This is possible when using ftp to save files.
6) Why are Tesla's the majority of cars without a front plate?
Box with side removed:
View before mounting box at corner:
The camera adjustment have been to set the video to H.264 and CBR = 10240; set the shutter to manual, 0-1 mS; enable FTP for image storage off camera; add it to Blue Iris; and set the capture area to a diagonal stripe. No optimization has been done.
Results have been excellent so far. No big problems even at night. All of the following images were sent by the camera to a computer, either via ftp or by the browser automatically downloading it.
Dirty plate:
Seems to have problems with speciality plates - tries to read the small letters.. Also has problems with symbols that are on some vanity plates - they are skipped.
Another interesting misread case:
Biggest problem I have is some plates are truncated at the edge of the FOV.
This is despite moving the capture area in from the edges.
I have done nothing except set the shutter speed. No optimization at all so far. Here's a night shot:
I believe this is a paper plate - I have never been able to capture one at night before.
Another paper plate, late in the evening. I do wish the camera would select better frames.
A Tesla with no front plate:
A few comments:
0) The camera should do a better job of selecting which frame to analyze.
1) Most of the web interface seems to work with any browser but the search function asks for old versions of Firefox or Chrome so that an extension can be loaded. Really??? That should have vanished years ago.
2) Under Settings --> ANPR --> LED Screen - what is this?
3) Under Setting --> Camera --> Image --> General - what does the default environment setting do?
4a) When looking at the camera from a browser, it started downloading the files to the computer using the storage path in 4b)..
4b) Under Setting --> Storage --> Storage Path - why is there no way to use the %09 macro to include the license plate in the filename? This is possible when using ftp to save files.
6) Why are Tesla's the majority of cars without a front plate?
Timokreon
Getting comfortable
Perhaps Wildcat1 can confirm, but isn't the plate frame able to analyze 1-3 frames for licensed plates and 1-10 for unlicensed? Perhaps change that setting to 2 or 3, which should be the 2nd or 3rd frame analyzed.
I hope this works. I changed Settings --> ANPR --> Snapshot settings
Snapshot quantity from 1 to 2 (max)
Frames to Output Licensed Vehicle Snapshot from 1 to 3 (max)
I should know tomorrow if this works.
Snapshot quantity does not seem to do anything except send an extra picture. The license plate cutout does not change between pictures and does not necessarily correspond to either picture. In the example below, the plate is not visible in either picture while the plate is correctly shown and identified. All photos are sent via ftp to my computer.
Here is a paper plate. I am quite surprised that it was correctly read.
And here is acement truck. The tilt of the plate seems to have been too much.
If the camera can't find a real plate, it seems to interpret anything in the general area.
These two are the only misreads out of about 30 cars since last night.
Here is a paper plate. I am quite surprised that it was correctly read.
And here is a
If the camera can't find a real plate, it seems to interpret anything in the general area.
These two are the only misreads out of about 30 cars since last night.
Here's an example where the camera capture a good license plate in one of its two images, included the good capture in its plate image, but returned the truncated plate.
Since I have been capturing two images sometimes the camera returns the first image (ftp) messed up. Never happened when only sending one image.
Here's an example of what happens with a very poor viewing angle - it missed the "1". I turned on adding the confidence level but will probably turn it off soon.
Sometimes the camera interrupts lettering on a vehicle for the plate. Also, note that the smaller size and worse angle causes it to misread some letters.
I do not plan on posting anymore daytime plates - the camera did not make any errors reading plates unless it decided to use a plate that was at the edge of the FOV, the plate symbols or small lettering, or if something was obscuring part of the plate. No motorcycles have passed by so far. Less than ideal viewing angles and distances from other setups will be interesting tests by others.
Since I have been capturing two images sometimes the camera returns the first image (ftp) messed up. Never happened when only sending one image.
Here's an example of what happens with a very poor viewing angle - it missed the "1". I turned on adding the confidence level but will probably turn it off soon.
Sometimes the camera interrupts lettering on a vehicle for the plate. Also, note that the smaller size and worse angle causes it to misread some letters.
I do not plan on posting anymore daytime plates - the camera did not make any errors reading plates unless it decided to use a plate that was at the edge of the FOV, the plate symbols or small lettering, or if something was obscuring part of the plate. No motorcycles have passed by so far. Less than ideal viewing angles and distances from other setups will be interesting tests by others.
Robert G.
Getting comfortable
I have made a decision on my cameras to use the Intelligence custom algorithm settings. I am using the current plate format(s) for my area, I miss any personal plates, motorcycle, trailers etc. but I find it keeps my false positives down (way down).
I am capturing about 3,000 - 4,000 plates a day and want to keep the data as clean as possible.
I am using this:
p replace=I->1
p replace=O->0
p pattern=EENNNE
p pattern=NNNEEE
p pattern=EENNNN
p pattern=NNNNEE
p pattern=EEENNNN
We do not see massive numbers of vehicles from other places, I know in the USA there always seems to be LOT of plates from other states around so this solution may not work great in your area.
I have been looking at how the LPR-437 does at night and am seeing problems with it not triggering on obvious cars passing through. I was going to add a couple of videos but BI is not exporting correctly. BI seems to only be showing key frames.
I compared BI motion triggers with the output from the LPR-437. Any time BI or the camera showed a trigger I checked to see if the other one triggered. Last night it captured 47 cars/plates, missed 17 times, and caught 3 cars that BI did not. I had BI setup with very lax motion triggers so I am not surprised that it missed a few times. I am extremely surprised that the camera did not trigger on each of the 17 times it missed.
Here are a few example BI snapshots from the LPR-437. In each case BI triggered on motion but the camera did not trigger. I see no reason why any of these should not have resulted in a good plate/vehicle capture
Any ideas? Anyone else missing vehicles at night? I am hoping that @Wildcat_1 will have some ideas in his nighttime review.
I compared BI motion triggers with the output from the LPR-437. Any time BI or the camera showed a trigger I checked to see if the other one triggered. Last night it captured 47 cars/plates, missed 17 times, and caught 3 cars that BI did not. I had BI setup with very lax motion triggers so I am not surprised that it missed a few times. I am extremely surprised that the camera did not trigger on each of the 17 times it missed.
Here are a few example BI snapshots from the LPR-437. In each case BI triggered on motion but the camera did not trigger. I see no reason why any of these should not have resulted in a good plate/vehicle capture
Any ideas? Anyone else missing vehicles at night? I am hoping that @Wildcat_1 will have some ideas in his nighttime review.
Timokreon
Getting comfortable
Yes, I miss a few plates here and there. I'm eagerly awaiting wildcat's night review to see what settings he has. I might play around with sensitivity a bit as well to see if that changes things.
I will say, during snow storms the ability to capture plates is greatly diminished. But to be fair, I'm trying to capture plates at around 80 feet... not Dahua intended target I'm sure.
I will say, during snow storms the ability to capture plates is greatly diminished. But to be fair, I'm trying to capture plates at around 80 feet... not Dahua intended target I'm sure.
Don't worry I haven't forgotten about this thread or the night video + the DM's just finished up a revisit comparison of the A180 to Hik vision and will be coming back to this one as soon as next week. Thanks
just finished up a revisit comparison of the A180 to Hik vision and will be coming back to this one as soon as next week. Thanks@Timokreon - I found during the recent rains, especially at night, that the camera missed a lot of plates/cars. But when it did trigger the results were still very accurate.
Found another case where the camera responds in a less than ideal way. Today a car happened to be parked on the far side of the street within the FOV. Any motion seemed to trigger and the same plate was returned many times. When another car passed in front I generally received captures for both the moving and parked cars. It would be nice if it would ignore the parked car except on arrival and departure, ignoring the static car in between. More worrisome is that sometime it only captured the parked car and ignored the moving car.
The red car's plate was not capture, just the background truck. The motion of the car triggered the camera, but it found the parked truck. Before this there was a long string of the gardeners moving and their truck plate being returned.
I set the retrigger interval to 1 second so that a tailgating car could be captured. Works great. It also works with two cars passing in opposite directions as shown in the following two captures.
This little girl seems to be operating an unlicensed vehicle
Finally, I'm amazed that this plate of a parking car was read correctly.
I like the results with this camera enough that I have ordered a second one.
Found another case where the camera responds in a less than ideal way. Today a car happened to be parked on the far side of the street within the FOV. Any motion seemed to trigger and the same plate was returned many times. When another car passed in front I generally received captures for both the moving and parked cars. It would be nice if it would ignore the parked car except on arrival and departure, ignoring the static car in between. More worrisome is that sometime it only captured the parked car and ignored the moving car.
The red car's plate was not capture, just the background truck. The motion of the car triggered the camera, but it found the parked truck. Before this there was a long string of the gardeners moving and their truck plate being returned.
I set the retrigger interval to 1 second so that a tailgating car could be captured. Works great. It also works with two cars passing in opposite directions as shown in the following two captures.
This little girl seems to be operating an unlicensed vehicle
Finally, I'm amazed that this plate of a parking car was read correctly.
I like the results with this camera enough that I have ordered a second one.
I now have the second 437-LPR set up looking the opposite direction down a residential street. Problem is that there are usually cars parked along the street in this direction whose plates get read instead of the moving car(s).
Last night while is was raining it kept reporting the plate from a car that was parked. It seems that it should remember that the plate has not moved and not report it every minute 30 times in a row. It should also know where the trigger motion occurred and look in that region, not some area where no motion occurred.
While a car is moving through the FOV it again reports all the parked cars and usually, but not always, catches the moving car. When the moving car appears, it seems to read the first image where the plate enter the capture area. It then proceeds to read the plates of all the parked cars.
It should capture a full plate with the greatest confidence. Even better would be if it would combine the reading as the plate crosses the capture area. This would eliminate partial plates at the edges, ensure that the best readings are used, and might let a full plate be read that is partially obscured by a trailer hitch or bike rack or even one of those anti plate reading covers.
I tried reducing the capture area so it did not include the parked cars. Did not help - the parked cars are still read. Does the capture area just affect the trigger? Sure doesn't affect the plate read area.
Also, when there are gaps between the letters the reading should place a blank between the characters. Likewise, when a character is obscured or poorly read it should return a question mark. And when the letter size changes dramatically it should recognize that. This would prevent errors when reading handicap plate or when someone puts a US plate over a Euro type plate.
Last night while is was raining it kept reporting the plate from a car that was parked. It seems that it should remember that the plate has not moved and not report it every minute 30 times in a row. It should also know where the trigger motion occurred and look in that region, not some area where no motion occurred.
While a car is moving through the FOV it again reports all the parked cars and usually, but not always, catches the moving car. When the moving car appears, it seems to read the first image where the plate enter the capture area. It then proceeds to read the plates of all the parked cars.
It should capture a full plate with the greatest confidence. Even better would be if it would combine the reading as the plate crosses the capture area. This would eliminate partial plates at the edges, ensure that the best readings are used, and might let a full plate be read that is partially obscured by a trailer hitch or bike rack or even one of those anti plate reading covers.
I tried reducing the capture area so it did not include the parked cars. Did not help - the parked cars are still read. Does the capture area just affect the trigger? Sure doesn't affect the plate read area.
Also, when there are gaps between the letters the reading should place a blank between the characters. Likewise, when a character is obscured or poorly read it should return a question mark. And when the letter size changes dramatically it should recognize that. This would prevent errors when reading handicap plate or when someone puts a US plate over a Euro type plate.
Robert G.
Getting comfortable
Have you tried to apply a shield area, the shield area button is on the same page as your screenshot here. I believe you apply that to prevent any captures of vehicles in that area.
The capture area is not the area to capture plates, but where to 'snap' the photo of the moving car. You could for example, make your capture area 1/2 the size that it is currently, so it will capture the plates & vehicles when they are in the bottom 1/2 of the image making them larger for your photos. If you set the capture area at the top portion of your field of view it would snap the photos in that capture area, making your plates smaller in your captured images.
Thanks, I had not considered or realized how shield areas might work. I will try both shield and reducing the capture area later this week. One problem is that I expect the plate from trucks will overlap with the plates from parked cars. Also when the parked cars are not on that side moving cars tend to be closer to the curb. It will require experimenting to get a good balance between missing plates and redundant parked plates.
I am hoping @Wildcat_1 will have more information on how to set up this camera when he finishes his review. It is a strange beast and a lot of controls are new or don't work the same as other cameras. Right now I am trying to get a feel for how this camera works and am expecting the final setting to be zoomed all the way in.