Sorry, still a little new with the NVR.
when you say try using the AI on recorder, do you mean the search area? if so that's what I was asking about..
Is it possible to get a notification when a particular vehicle passes my home?
- Thread starter ricklndn
- Start date
when you say try using the AI on recorder, do you mean the search area? if so that's what I was asking about..
bigredfish
Known around here
I have a much older NVR that doesnt do AI detection by itself but does interpret AI data/attributes from the camera (5216-16P-4KS2E)
I can choose multiple attributes and sort results
(camera is a 5442-Z4 (S3)
(*One of the reasons I dont feel compelled to upgrade to the new Ei or Xi NVRs is this, plus I have no need in a residential setting for AcuPic or doing AI by recorder, as its typically less accurate than by camera anyway)
When setting up the camera... If you go to EVENT / AI SETTINGS / then VIDEO METADATA / it will give the option for AI by Recorder or by Camera.
You can do that on the 608 as well. Not sure why he isn't able to choose vehicle type.
Mark_M
Known around here
Yes you can get notifications for a vehicle. Dahua NVRs do not use the term "whitelisted" and "blacklisted" now.
As far as I know, you need a Dahua ANPR camera to get only 'block listed' vehicles notification. Using the video metadata method would give a notification for every vehicle detected.
Video metadata method:
You need a 5 series or higher NVR.
If you have a 5 series NVR with either "-I", "-E/I", "-E", "-XI", "-AI" (and some other versions) then it probably has on board 'Video metadata'. This will not work if you have a "NVR5x-4KS2" NVR. Check your NVR's specifications.
Go into settings > AI > Parameters > Video metadata.
Select the camera you have. Which needs to be properly aligned and setup for capturing plates. Keep in mind you need decent FPS for fast moving traffic.
Enable "VehicleDetect" rule on the camera.
The NVR will attempt to do ANPR, alongside detecting a vehicle type, color, etc.
My camera is not setup for ANPR (too wide, wrong angle, wrong type of camera). I use video metadata for car counting. I have a different ANPR camera.
Occasionally it picks up a plate, and almost got it right
The actual 'ANPR' with general/block/allow list requires a Dahua ANPR camera. 5 series NVRs state 'ANPR by camera'.
In DMSS app >Device details > multi alarm subscription > Other alarms > License Plate Recognition > Select your video channel.
With only the video metadata type on my system I do not see a method to only notify on block listed plates.... because there is no block or allow list when using video metadata.
As far as I know, you need a Dahua ANPR camera to get only 'block listed' vehicles notification. Using the video metadata method would give a notification for every vehicle detected.
Video metadata method:
You need a 5 series or higher NVR.
If you have a 5 series NVR with either "-I", "-E/I", "-E", "-XI", "-AI" (and some other versions) then it probably has on board 'Video metadata'. This will not work if you have a "NVR5x-4KS2" NVR. Check your NVR's specifications.
Go into settings > AI > Parameters > Video metadata.
Select the camera you have. Which needs to be properly aligned and setup for capturing plates. Keep in mind you need decent FPS for fast moving traffic.
Enable "VehicleDetect" rule on the camera.
The NVR will attempt to do ANPR, alongside detecting a vehicle type, color, etc.
My camera is not setup for ANPR (too wide, wrong angle, wrong type of camera). I use video metadata for car counting. I have a different ANPR camera.
Occasionally it picks up a plate, and almost got it right
The actual 'ANPR' with general/block/allow list requires a Dahua ANPR camera. 5 series NVRs state 'ANPR by camera'.
In DMSS app >Device details > multi alarm subscription > Other alarms > License Plate Recognition > Select your video channel.
With only the video metadata type on my system I do not see a method to only notify on block listed plates.... because there is no block or allow list when using video metadata.
bigredfish
Known around here
I think you meant faster shutter speed instead of FPS to capture faster moving vehicles? I capture clean images at 8-15FPS depending on the camera.
Are you saying you can get notification just on a specific vehicle without ANPR, or in general you can get notified for every vehicle? If the latter, that would depend on how much traffic the OP gets if it is worth that many notifications.
I get about 400 per day, but as you say, I'm looking for a exact match, for example, a car that's white with a red door... So basically I would have to have a screen shot or copy of the vehicle of interest.
As of for now this is not possible.
Without a plate, I don't think an NVR is capable of that. With a PC and Blue Iris you could train a model with your field of view and just that truck, although I have not heard anyone try it for just that. You may still get some false triggers, but not a trigger for every vehicle.
I'm not really caring anymore for this feature, it would be nice, but I can go through the vehicles daily using the AI captures .. not bad .. just would be nice for a separate look of what I program to get notifications of interest.
I'm lucky that I'm able to search AI for color.. this definitely narrows down search
Mark_M
Known around here
Yes and no.
The camera angle to the plate when straight-on can get away with lower FPS.
My own ANPR camera is currently running at 50fps, auto shutter 1/500 to 1/100,000. Playing back video, a speeding car only gets captured in 20 frames.
Milesight sell an ANPR camera at 100fps too and most Tattile ANPR cameras are at least 75fps.
Without an ANPR camera: Only notification per vehicle detected going past. Including vehicles without a plate detected.
With an ANPR camera: I presume DMSS gives the option for block-listed notifications only (I don't have a Dahua ANPR cam to test).
The metadata search is not like facial recognition (database partition) or license plate recognition (database partition) hence there is no "database" to compare and notify from. Only tags attached to the video that the camera inserted during the video stream that the recorder can query. You cannot flag a vehicle based upon shape or color and get a notification. You could only get a notification if a vehicle was located within the field of view of the camera. Hence detection not recognition.
If you cannot get a shot of the plate ANPR would not work for you. That being said the EmpireTech IPC-T58IR-ZE-S3 (Dahua IPC-HDW5842T-ZE) & EmpireTech IPC-T54IR-ZE-S3 (Dahua IPC-HDW5442T-ZE) do not support ANPR anyway. Since the base camera for Dahua doesn't support it, Empire's won't without heavy modification which requires NREs (non recurring engineering fees). That being said even if Empire Tech was willing to pay for it, both of those cameras only support SMD 3.0, which is deprecated.
AcuPick
------
Uses deep learning algorithms and works with back-end devices to accurately match targets, such as people, animal, and motor vehicles, and search through live and recorded videos to quickly locate targets.
Video Metadata
------
Motor vehicle, non-motor vehicle, face, human body detection; track; snapshot; snapshot optimization; optimal face snapshot upload. Motor vehicle attributes: vehicle type, vehicle color, vehicle logo and other attributes : seatbelt, smoking, calling. Non-motor vehicle attributes: type, color, number of people, top type and color, hat. Human body attributes: gender, top/bottom type and color, bag, hat and umbrella. Face attributes: gender, age, expressions, glasses, face mask and beard.
Notice how both functions are for searching video and not comparing against a database for recognition. Unlike face recognition where you can utilize "stranger" mode or in your case "Known person" mode where you could take a snapshot of their face, add them to a database, then flag that face to generate a notification with a timestamp. Same with a vehicle license plate.
You'd be better off streaming one of the cameras RTSP feeds to a Raspberry Pi 5 16GB or ODROID-M2 6GB then use YOLOv5 (Ultralytics) for object recognition that you can train. Just go to chatgpt or grok and type in "How can I train a raspberry pi watch video from an RTSP stream and use something like ultralytics for object recognition?"
Here is a guide
------
OR you could just set your cameras to use SMD 3.0 vehicle detection and have it immediately auto email you a snapshot each time a vehicle is detected that you can review at a glance.
I really appreciate the effort it took to put everything into perspective, thank you!
At the end of the day I just do an AI search based on vehicle and color, it displays what I need.