License plate/automatic number plate recognition is increasingly used in security as more options and analytics advance, but what is still missing and what are the improvements integrators hope to see?
Related, see Favorite and Worst LPR/ANPR Providers.
In this report, IPVM analyzes the 6 main improvements that 150+ integrators told IPVM are wanted.
Executive *******
*********** **** ******** ****** ************ ** read ********, ************ *********** *** ****, smaller **** *******, *********** *************, *** cost. **** ************ ********* ******* ****** angles, *** *****, *** **** ****** as ******. **** *** *** ****** integrators *******:
Reading ********
*********** **** ****** ************ ** **** accuracy, ********* *** ***** **********, ******* angles, ****** *****, ********* ********, *** higher ******. **** **** ********* **** cameras **** ******* ******* ******* ********** on ******:
*** ********** ********* ****** ************ ** HD ***** ******* ****** **** ******** as ** *****:
* ***** *** ** *** ** issue ***** *** ****** ***** *** visibly *** *** ******. ** **** as ** ***** ** ******* *** the ********* **** *** **** **** was *** ***** ** **** ** could ***. ***** **** ** *** high-def ******* ** *** ** **** clients ***** **** **** ***** '******* plate ***********'
** **** *******, ****** ****** *** range **** ******** *** ******** *****, and ******* ******* *** ****** **** stacked ********** *** ********* ******. *** our*** ********** **** ******* ** ***** *** other ******.
Physical ****** & **** ******
******* *********** ****** **** ****** ******** requirements *** **** *********** ** ************, as **** ** ********** **** *******.
********* *** ****** ****** ********* *** bullet ** *** **** *******, ** these ****** **** ***** ***** *** the ****** ****** *** ****** ** illuminators **** ** *** *****. *******, they *** ** **** ***** ** vandalism *** ** ***** ****** *********, compared ** *****/*******, ***** *** ** preferable **** ******** ** *** ******* monitoring ***** *********/***** ** ***** *** speed ************.
Ease ** *** & *****
**** ********* **** *****'* **** ** improve ** ***** *** **** ** use *** *** *****:
Compatibility & ***********
**** ********** **** ***** ** *** enough ************* ** *** *****, **** many ******* *** ********** **** *****, via ***** ** *********:
Cost, ******, & ******** *****
**** *********** ******* ***/**** ****** *** too ********* *** ***** ** * lack ** ***** *** ***-***** ********.
**** ********** ********* *** ******* ********* sell *** ********* ** ******* ****, such *********'* ***** **************'* ***. ***** *****/***** *********** ****** *** be ***** **** ***** *** ****** for ~$*,*** *** *****, ****** **** will **** **** ** ***** ****** or ** ********** ****** ************ *** typically **** ******** ***** ** **** expensive ****** (*.*., ******** **** ********, make/model *********, ***** *********).
Database & *********
**** *********** ********* ***** **** *** better ******** ********** *** *********** **** law *********** *********:
Few ********* ** *****
***** *** ***** ** ********* * top *****, **** *** ********** ********* wanting *** ** *** *****:
***** *** *******. *** ***** ********* right *** ***’* **** * ********* LPR ******. ***** ********** **** *** reliable ****** **** ***** ******* ** they ***.
*********, ***** ********* *******,*******, *********** *** ***** ******** *** *******, but ******* ****** ********* ***** ***** bullet ****** *** *** ***********. ****** offer ***** *** ***** *********** ********, such *******,********'* ***, ** ********** ****** ***********.
**** ********* *** **** ***** **-******* server *****, ****** **** ******** ***** options **** ****** **** ****** ** the **** *** *****, **** ************ **** ************'* ******** *****. ** ****** ***** ******* ** continue ** **** ** ******** ** cloud/SaaS ********* ******* ****.
Interesting report. Seems like this is based on what the integrator knows not what products are on the market.
One improvement I would like to see is the ability for AI/ALPR to detect if there is a vehicle on the scene and then check to see if there is a plate on that vehicle. My understanding is this takes a lot more processing power most systems do not do this.
Also would be nice for more gate control solutions if you could require the color, make, model plus ALPR data for the gate to open.
Wow... Lets break down a few issues here:
1- Accuracy is based on distance and angle of attack, Period. You can't have both!
This takes me back to the old VCA days. The camera can only see what your eyes can see. VCA, Machine Vision, or AI, whatever the latest trite trend name is in the industry the technology is based on pixel analysis. Yes, we have more pixels than we did in the 80's and we have more processing power to analyze those pixels, but the baseline is the same, you can't read it? the LPR won't read it either.
When you introduce steep angle of attack the license plates are no longer rectangles, the fonts change and the spacing between them decrease. Which is why most LPR manuals insist on 30 degrees.
2- Distance reading is pure optics. LPR can read from a mile away if you have the proper lens, BUT again you are sacrificing accuracy as you add shake, pollutant particles and bad weather such as fog, rain and snow to the equation.
3- Cloud based LPR is a bad idea all around! If you use LPR for access control you will piss off the residents as they wait by the gate. If its used in high-speed application by the time you get the result the vehicle is gone. SPEED is a major factor in the usefulness of any LPR system.
4- Compatibility and Integration with ONVIF is meaningless. Stop thinking about it as a security camera in which you want to stream video! it is not... LPR main purpose is to capture a few quality snapshot which will provide an accurate read. You want to stream video from the road scene? use a separate HD camera with a wider view. the LPR camera performs best when it focuses on one lane.
Which gets us to the multi-lane requests
5- Multi-lane reduces accuracy of the reads, AND of the identification of the vehicle. Multi lane LPR relies on VCA to first capture the vehicle, capture the plate area and then run the OCR engine. The process decreases accuracy the more lanes you want to cover. It is completely useless for access control.
Finally, the best way to identify a vehicle in the scene is what we do in access control applications. We tie the LPR to a vehicle induction loops. We use the camera Alarm Input to trigger the LPR/OCR engine. Vehicle Induction Loops are used in traffic lights all over the world and are the most reliable tool for vehicle presence.
One correction. the word "based" is inappropriate here. Accuracy is affected by distance and angle. It is based on many other factors I will not delve into here.
Huge Kudo's to IPVM for this. Extremely helpful and as a analytics manufacture who is being requested to provide LPR almost on a daily basis this helps firm some customer opinions we have heard.
Cost and Accuracy are a major concern for our customers where LPR is "nice to have." If it's required, cost isn't a concern. We also find that having this entire workflow delivered on an edge device without requiring specialty cameras is very attractive.
From the analytics perspective. LPR is hard, especially in the US, the variety of plates, numbers and configurations across the US is exponential. High Accuracy requires high compute and constant updating of the model to adapt to new scenarios in ambient camera style situations.
High performing dedicated LPR cameras typically sell for thousands of dollars each, such as Genetec's Sharp line or Motorola's L5F. Lower speed/lower performance models may be found from Hikua and others for ~$1,000 and under, though they will work only at lower speeds or in controlled access applications and typically lack features found in more expensive models (e.g., multiple lane coverage, make/model detection, speed detection).
Everyone wants high performance at value prices, most consumers watch too much TV and have no what it takes. There are many valuables in LPR and analytics in general that the reseller fails to set the proper expectation for the user. I would wager for most applications a glorified OCR engine would work great as most people are doing forensic searches over time, sell it like it is a "Search Tool" You want watch lists and other fancy stuff then fork over the $$$ and pay for those advanced features that deliver.
Take a look on the traffic signals where you live, I will bet you see more and more L5F ANPR cameras installed. The Sheriff where I live just sits in between two signals about 1/2 mile apart and waits for hits coming from Casino Corridor. Like shooting fish in a barrel.
ANPR at traffic speed is a requires fast cameras and paired lighting. I am curious how many of the service providers on IPVM have experience with Transportation Surveillance and Monitoring including ANPR at speed.
