LP snapshot with motion

[cjowers]

Hi All,
My setup captures LPs great on continuous recording, but i'd also like to have the snapshots (for playing around with AI and for longerterm storage instead of videos), but I have some difficulties with the variability on the timing of the jpeg capture (which is currently based on BI motion trigger).

The main components affecting this seem to be:
-car direction (whether back or front plate),
-day or night (headlights trigger before vehicle),
-vehicle speed,
-others?,
I'd like to improve the consistency just a tad - my goal would be to get most LP captures 'in frame.' right now it's probably only 60%. Has anyone worked on these issues before? Any simple solutions I could I do?

only thoughts coming to mind so far....
using a BI night profile (trigger later / after headlight motion)?
more jpeg captures per car pass? (might increase computation / storage, but could also give secondary vehicle ID visual confirmation?)
IVS trip setup including direction of motion? (added delay for rear plates)
zoom out fov (I'd probably lose some of my nighttime accuracy, as I've played with this already)
other cheap sensor methods
???

Thanks

 

[wittaj]

I have found that IVS doesn't work for the zoom needed as the camera needs time to recognize the motion and did it cross a zone. If the vehicle is in the frame for less than a second, IVS probably will miss it.

It takes some trial and error in BI, but setting up a zone works well. You will have to run a separate night profile and make the target a lot smaller as most will only see head/tail lights and the plate on the image. My target at night is about the size of the plate.

Here is my setup and you can see from the alert images, I am getting them. In my case, it was draw a zone where I did so that the cars going right to left, I catch the front plate, and cars going left to right, I catch the back plate. It is some trial and error to get the FOV and zone line drawn to where you get them, but you see the idea in how to get it to trigger an alert with the vehicle in the frame. I do zoom in at night as this FOV for nightime was just a little too wide, but it is fine for daytime.

And then if you only wanted to catch vehicles going in one direction, simply add another zone to account for that.

 

[biggen]

That's a good way to do it @wittaj. I've been using my overview camera to trigger a group alert for which my LPR camera is a sole member of the group. Its works fine but generally the car is a second or so outside of the LPR alert thumbnail frame since the overview camera has a larger FOV and triggers the group alert before the LPR camera can actually see the car. I may give what you did a try today and see how it goes.

I agree that IVS isn't very good for LPR. It wasn't triggering for me when I tested it for several days.

 

[cjowers]

Cheers,
I'll give this a go. Makes perfect sense And my images look very similar to yours, maybe a little more head-on, so I'm expecting good things.

And for the night profile, would I just add a bit of time delay to the jpeg capture? Since moving the zone (to accommodate the headlights at night) would only work for 1 direction of traffic?

Thanks again

 

[wittaj]

What I found was that the tailights were still bright enough to trigger even at night. But you have to make the object size A LOT smaller since the image is black and all you see are the head/tail lights and plate.

 

[cjowers]

I'm getting the hang of it, but i've noticed a few things:
-my images aren't as dark (my cam is setup to also pickup people on driveway at night, so perhaps it's not ideal for this level of control)
-the car headlights light up the road surface considerably more than yours, and are usually what triggers the cam first, since they hit the zone first. my motion sensitivity isn't cranked up like yours, just moderate. I imagine it is because my images aren't dark-dark, instead they're only dark enough to make out the plate numbers.

 

[wittaj]

Yeah, not having the image dark enough would certainly cause those issues. I have mine dedicated just to plates, so it is optimized to just see head/tail and plate at night. Occasionally under the right rain conditions I may get just a blip of that light shine on the pavement.

What you can do is watch a few plates and see how long it takes from light shine until vehicle in frame and then adjust the make time accordingly. Maybe something like 0.2 at night is all you need. Obviously that will be vehicle speed dependent, but you could certainly get it a lot closer that way.

 

[cjowers]

DeepStack OpenSource and Custom Object Detection Releases

Hello everyone. We have some exciting news to share today. DeepStack is now open source on Github, johnolafenwa/DeepStack: The World’s Leading Cross Platform AI Engine for Edge Devices (github.com) This is super exciting for us, and we are looking forward to developing DeepStack in partnership...

forum.deepstack.cc

there is a new deepstack version out that allows for custom models. Anyone using this yet to detect plates in their still images? Seems like a good way to reduce false hits / filter out just the plates for further recognition or storage, object coordinates available for cropping too