A set of services written in Go for running OpenALPR on the Raspberry Pi. These are written in such a way that they add missing functionality to the open source version of OpenALPR (e.g. multiple detection threads).
The application is broken up into microservices, so that a cluster of Raspberry Pis could be effeciently used to increase the detection rate on a single camera. For example, you might have a very busy roadway, and thus want more detection threads running by horizontally scaling the number of plate_detector instances across many Raspberry Pis.
postgres -D /usr/local/var/postgres
beanstalkd -V
cd plate_detector && . lpr.env && go run plate_detector.go
cd uploader && . lpr.env && make run
cd watcher && . lpr.env && go run watcher.go
cd watcher/testdata
./create_jpgs
Use scripts/schema.sql to create the schema for the main PostGres DB.
LPR-Raspi relies on the excellent Motion project, which does motion detection on video feeds.
The folder /motion_docker allows you to run Linux motion daemon on OSX. There are make commands in there that build and run the container.
It's possible to create a MJPG stream of a local video for testing purposes. First grab sample video and resize it:
ffmpeg -i sample.avi -vcodec libx264 -preset slow -profile high -vf scale=768:432 432p/sample.avi
From VCL create an MJPEG stream to Raspi:
vlc \
--loop \
-vvv \
sample.avi \
--sout "#transcode{vcodec=MJPG,fps=5}:std{access=http{mime=multipart/x-mixed-replace;boundary=--7b3cc56e5f51db803f790dad720ed50a},mux=mpjpeg,dst=10.0.0.5:8082/go.mjpg,delay=0}"
And to Motion on localhost or Docker:
vlc \
--loop \
-vvv \
sample.avi \
--sout "#transcode{vcodec=MJPG,fps=5}:std{access=http{mime=multipart/x-mixed-replace;boundary=--7b3cc56e5f51db803f790dad720ed50a},mux=mpjpeg,dst=127.0.0.1:8082/go.mjpg,delay=0}"
Then in browser:
http://10.0.0.5:8082/go.mjpg
You can try specify bitrate / FPS, but prefer just using fps.
vb=10000,fps=10
vb=5000,fps=5
vb=2000,fps=2
The plate_detector app runs openalpr.RecognizeByFilePath and detects plates in the images using OpenALPR.
The Uploader service picks events off the beanstalk queue, creates crops and thumbnails of the JPG event image using GraphicsMagick, and uploads the results to the central PostGres DB in the cloud, and Amazon S3.
For GraphicsMagick integration on a Mac:
brew install homebrew/versions/giflib5
export CGO_LDFLAGS="-L/usr/local/opt/giflib5/lib"
export CGO_CPPFLAGS="-I/usr/local/opt/giflib5/include"
go get github.com/rainycape/magick
Local DB for the deduplication:
brew install postgres
postgres -D /usr/local/var/postgres
createdb `whoami`
psql
Install the schema as per uploader/schema/uploader.sql
Beanstalk is the backbone queue (message bus) that the services use to pass messages.
- https://godoc.org/github.com/kr/beanstalk
- http://kr.github.io/beanstalkd/
- https://github.com/kr/beanstalkd
To start on OSX in foreground:
beanstalkd -V
# Get some stats
echo -e "stats\r\n" | nc localhost 11300
echo -e "stats-tube detection_events\r\n" | nc localhost 11300
echo -e "stats-tube motion_events\r\n" | nc localhost 11300
On Raspi (systemd):
systemctl start beanstalkd
The plate recognition is performed using OpenALPR integration:
- https://groups.google.com/forum/#!forum/openalpr
- https://github.com/openalpr/openalpr
- http://doc.openalpr.com/
Installing ALPR on OSX:
brew tap homebrew/science
brew install openalpr
Running OpenALPR:
alpr -c eu -j 01-20160609180828-02.jpg | jq '.'
Compiling ALPR on Raspi can be done using the lpr-deps repo.