Capture stills with Raspberry pi?

[Marble68]

I am just getting into LPR projects - and I have a 2MP LPR camera hot off the boat from China. ha
2.0mp Traffic Camera Hd Lpr Camera Car License Plate Recognize Lpr Dash Camera - Buy 2.0mp Traffic Camera,Hd Lpr Dash Camera,Car License Recognize Camera Product on Alibaba.com


It provides an RTSP stream - and I setup FFMPEG to grab frames and write them to a ram drive.

It seems to work ok - but as cars go past it sometimes "ghosts" them, or misses them entirely.

My goal for this project is to put the camera on axis, and using a radar module, take a photo of every vehicle coming down the road.

I figured I'd probably use openALPR for post image processing - but is that the way to go? There's also OpenCV for real time object detection and tracking - is that what other people use?

Would I do better to consume the video stream and process it live and save out pictures that are clear?

My current solution has FFMPEG writing frames out to a ram drive - then another job cleaning up old images - and finally, another job that copies the images based on radar input.

At the end of the day - my goal is a clean picture of every vehicle that passes my camera with human readable license plate - and hopefully clear enough for something like openALPR to process with reasonable accuracy.

I'm running the camera at 1280x1080 30fps - but the PI is only grabbing about 5 to 7 fps (with ffmpeg).

 

[richardgohth]

set the shutter to 1/1000 ( exposure time 1msec ) or faster

 

[carbonita]

I have a raspberry pi zero which captures stills when confidence number is high for faces. It can do (maybe) one still every 3 seconds. I used the Google AIY kit for the parts, and modded code for face stills and email. Separately, google vision service has a demo, where one can upload images to assess. I figure google is at least competitive with OCR.

 

[Marble68]

set the shutter to 1/1000 ( exposure time 1msec ) or faster

That makes sense, I see this mentioned in other threads. I'll try and determine how this is done. Thanks.

 

[Marble68]

I have my camera set to maximum frame rate, etc. - I was reading on an openALPR thread about dropping the frame rate to match the platform processing it.

Has anyone else had luck with this approach?

I was thinking an openALPR license would be a good idea - but they want a $1000!!!!!! That's insane.

 

[Marble68]

I have a raspberry pi zero which captures stills when confidence number is high for faces. It can do (maybe) one still every 3 seconds. I used the Google AIY kit for the parts, and modded code for face stills and email. Separately, google vision service has a demo, where one can upload images to assess. I figure if google is at least competitive with OCR.

Those look cool - I'll have to play with one! Thanks!