Maybe someday there can be a feature to catch stuff like this. I am not sure if this is legit or not, but I have noticed a lot of paper plates that look sus.
Full ALPR Database System for Blue Iris!
- Thread starter algertc
- Start date
TheWaterbug
Known around here
PSA, if you want the database container to start automagically after boot, you can set Docker to start automatically on login:
and then use netplwiz to enable auto-login:
No regedit required!
You may have to turn off Windows Hello in Sign-in options to enable that checkbox in netplwiz:
Note also that you can't turn off Windows Hello sign-in from a Remote Desktop session; you must do this from a real KVM session, or from a KVM-over-IP session.
and then use netplwiz to enable auto-login:
No regedit required!
You may have to turn off Windows Hello in Sign-in options to enable that checkbox in netplwiz:
Note also that you can't turn off Windows Hello sign-in from a Remote Desktop session; you must do this from a real KVM session, or from a KVM-over-IP session.
What exactly do you mean @PeteJ ? I'm all for more ways to catch threat actors, but I'm not sure I understand what you're saying. Distinguishing between paper plates is well beyond what we are currently able to do, but the ALPR is getting very good, and we will be able to detect most plates very soon.
It's more wishful thinking, but it would be great to be able to tie the plate to some form of vehicle description (type, make, model, color, etc) and be able to run a query to see if that changed. In the case above, the license plate of the vehicle changed--maybe the original plates were stolen, or maybe they've given up their personalized plate, who knows, but it is a little odd.
There are lots of paper plates where I am, and apparently, there are 2 types:
I am picking them up just fine in the daytime, but I get the feeling there are a number of fake ones and some of them are changed from time to time? Would be great to be able to tell if a vehicle's plate changed so that I can pay more attention to these vehicles.
I realize my use case is not common; I am in a high crime area--2 murders on my block within the last 4-6 months. The most recent one drove right by my cameras, which is when I started adding LPRs. What does this have to do with changing plates? I think there are a few stolen vehicles with paper plates.
I probably have about 4K valid plates in my database, of the ones I've tagged, the top 3 are:
Anyway, I am not requesting this as a feature, I believe there are more useful things that can be added that would benefit the community at large more than this feature.
I might be thinking of the wrong person, but I thought @Robert G. had something going where he could tell if plates were being changed on cars.
He will probably chime in soon, but if I remember correctly, he had his own database program he set up that could do this. Maybe we can add this as a feature here sometime
I would love to hear any way to distinguish fraudulent plates. Anything like that would actually be at the top of my priority list. I live with my parents in a decently nice neighborhood, but there are still quite a lot of thefts. I'm all for anything that might catch any burglars.
I've been working on the ALPR module integration and talking to @MikeLud1. I think we are making very good progress, all things considered. I've been quite busy this week, but we will definitely have something upgraded and very good soon.
TPMS iz also my main priority now that other bugs are worked out. I am going to try to build something to intercept and log TPMS and Bluetooth soon.
TPMS iz also my main priority now that other bugs are worked out. I am going to try to build something to intercept and log TPMS and Bluetooth soon.
While this might not be the easiest to do in the immediate future, I definitely see the value and would like to be able to flag those myself. Once we can get an improved recognition model working with the current situation, this could be an interesting thing to try to integrate.
I doubt there are many public datasets of paper plates in the USA, but that is sort of the beauty of this project anyways. The longer it exists, the more images will be captured and the greater the capability of the model will get.
@MikeLud1 was testing some state recognition a week or two ago, so this might be able to fit into that at some point. I definitely wouldn't dismiss it, it's just a tricky problem to address accurately.
Here is an example of what I assume is a fake temp paper license plate. You can clearly see the photo copy misalignment and an expiration date of August 2025. Per CA Vehicle Code 4456, Section 3, Part C, paper plates are valid for only 90 days. Also, temp plates do not have perm plate numbers, it's completely different.
Jeez yeah if that's paper/fake we're gonna need some smarter CV people to get on the case... I wouldn't personally have much confidence in developing anything to distinguish that...
It's possible that @MikeLud1 's state recognition might be able to grab the other parts and flag anything that wouldn't be possible in terms of date/time, but that still a pretty tough situation. The other hard part about it is that not all of the papers are reflective if they are fraudulent.
Again, the ability to aggregate data from all of the users is a huge advantage for developing something like this and is a great foundation, but I would honestly have no idea where to even start for distinguishing something like that. Even for a human, that really just looks like a real license plate to someone who doesn't look at a lot of plates.
Very open to any suggestions, and I have some substantial infrastructure available to develop any ideas on, but I'm far from an expert when it comes to to the nitty gritty classification like this.
Yeah, I think once we are able to train our own models, or submit images for training, we can supply temp plates from our state and maybe be able to have CPAI be able to distinguish the differences between "normal" dealer-issued temp plates vs perm plates. With this, maybe you can add a configurable feature to auto tag temp plates in the database that are more than X days old? If the temp plate is more than 90 days old in CA, it's expired, or maybe fake.
I'm working on the AI training. Turning out to be a bit harder than expected. Hitting some minor roadblocks in getting python model adjustments to work... Hopefully will be able to sort out soon.
My work kind of goes in cycles of downtime and extremely long hours, so I'm hoping to have another period of downtime to work on this. I want to see it work well, both for others and myself. I should have a nice period to develop it soon, but it's variable and has been very dynamic lately.
It's just hard to predict, as things change and really large travel-intensive jobs often hit me with no notice - sorry everyone for the delay...
Again, I really do enjoy working on this and it's not a chore at all. I really enjoy developing it and just want to see the BI ecosystem/community grow. I obviously want it for myself, but It's really cool to see others using it, too. I think about it often and am still very much on the case to improve all aspects of both the ALPR and object detection.
----
@PeteJ
That actually raises an interesting idea.
Right now, the tags are entirely up to the user, but I'm sure there are some that could help training. I'll look into this to see if there are any that could help improve annotation and try to place them nicely and separately into the UI for the users who actively watch their feeds. I recently learned from @MikeLud1 that day/night is already/automatically included in the AI result, but I'm sure there's lots we could do with a small set of standardized tags.
I've said this to a few others, but the potential to crowdsource data like this and create such a diverse training set is really exciting to me. One of the huge advantages that companies like Motorola have is that they have installations all over the world to train on. This is really the next best thing and could lead to a really phenomenal open-source model for Blue Iris users.
My work kind of goes in cycles of downtime and extremely long hours, so I'm hoping to have another period of downtime to work on this. I want to see it work well, both for others and myself. I should have a nice period to develop it soon, but it's variable and has been very dynamic lately.
It's just hard to predict, as things change and really large travel-intensive jobs often hit me with no notice - sorry everyone for the delay...
Again, I really do enjoy working on this and it's not a chore at all. I really enjoy developing it and just want to see the BI ecosystem/community grow. I obviously want it for myself, but It's really cool to see others using it, too. I think about it often and am still very much on the case to improve all aspects of both the ALPR and object detection.
----
@PeteJ
That actually raises an interesting idea.
Right now, the tags are entirely up to the user, but I'm sure there are some that could help training. I'll look into this to see if there are any that could help improve annotation and try to place them nicely and separately into the UI for the users who actively watch their feeds. I recently learned from @MikeLud1 that day/night is already/automatically included in the AI result, but I'm sure there's lots we could do with a small set of standardized tags.
I've said this to a few others, but the potential to crowdsource data like this and create such a diverse training set is really exciting to me. One of the huge advantages that companies like Motorola have is that they have installations all over the world to train on. This is really the next best thing and could lead to a really phenomenal open-source model for Blue Iris users.
I'm really looking forward to the training features. My camera setup might yield good data for training since I am using a dual lens setup. The color images are sometimes difficult to read, but the IR allows me to correct misreads with great accuracy. Hoping the color images with the correct labels would help produce better OCR results.
hopalong
Getting the hang of it
I'm using Vigilant/Motorola LPR cameras taken off police cars. They were only $100 on eBay, unfortunately all of them are sold out, otherwise I would buy a few more.
hopalong
Getting the hang of it
Is it one like this?
Vigilant Solutions VVR-10-925 LPR Camera with mounting bracket | eBay
Find many great new & used options and get the best deals for Vigilant Solutions VVR-10-925 LPR Camera with mounting bracket at the best online prices at eBay! Free shipping for many products!
www.ebay.com
As an eBay Associate IPCamTalk earns from qualifying purchases.
Is it one like this?
Vigilant Solutions VVR-10-925 LPR Camera with mounting bracket | eBay
Find many great new & used options and get the best deals for Vigilant Solutions VVR-10-925 LPR Camera with mounting bracket at the best online prices at eBay! Free shipping for many products!www.ebay.com
https://www.magnumelectronics.com/web/content/401204?unique=1cdc14b4d3a2a99445d1233aaafb2b3d552d147a
The 4th gen (VSR-42-9xx) is what I am using. It's POE and the controller is a Raspberry Pi module, so fairly hackable. It's just an rtsp feed, so works w/o much fuss with BI. The magic is how it handles the IR led array--there is an Arduino chip that handles the pwm and flickers the leds rapidly based on the input of a photocell and temp (I'm guessing, I can't figure out the logic) and the result is that the plates get picked up as very high contrast images, which is perfect for OCR.
The older generation might not be POE and the controller might some something else. I was thinking about buy one and cracking it open to see if it'd work. Their 5th gen and newer uses a Jetson Nano and does the OCR onboard, so I suspect that there is some method where it outputs the plate reading, but those are $3500, a little out of my reach.
I also picked up a bunch of these:
New Vigilant Solutions VSR-60-02MP1 IP Camera IPC262ER9-X10DU 24V 1.5A 50/60Hz | eBay
New Vigilant Solutions VSR-60-02MP1 IP Camera IPC262ER9-X10DU 24V 1.5A 50/60Hz Comes as shown with accessories shown. Fast and free shipping!
www.ebay.com
It's a 2mp on a 2.8 sensor, 1-10x zoom. It's basically a Uniview rebranded camera with slightly different firmware. It's not as good as the dual lens, but at $35, the value is hard to beat. This was taken at 90-100ft. I'm still testing them out, but they are probably going to be good enough for my needs.
As an eBay Associate IPCamTalk earns from qualifying purchases.