In this project, that was my bachelor thesis, a two-stage system is implemented for detection and recognition of Persian plates in urban traffic cameras. For the Plate Detection phase, that includes finding the location of plates in each frame of the test video, the Faster-RCNN method is used that outputs detected plates' bounding boxes. Next level, Plate Recognition, that predicts Id of the detected plates and actually behaves as an OCR module, includes two sub-levels: First, image segmentation and some low-level image processing algorithms are used to find characters' contours in a given plate. Then, a deep convolutional neural network(SqueezeNet) with 22 classes is used to determine type of each character.

Detection frequency 20-25HZ

Recognition time per character = 0.002s, that reduces overall frequency of the system to 13-17 HZ. Actually frequency of recognition phase depends on the number of detected plates in a frame.

1. First install forked version of py-faster-rcnn that includes some changes in configurations and files.
2. Install Caffe.
3. Install ROS(instructions can be found here).
4. Download the code from this repository:

git clone https://github.com/mammadjv/vehicle-registration-plate-recognition.git

5. Download the pre-trained file on plates dataset from here.
6. Change model paths in /path/to/vehicle-registration-plate-recognition/src/plate_detector/scripts/plateDetector.py, in line 36.
7. Change model paths in /path/to/vehicle-registration-plate-recognition/src/char_recognition/scripts/charRecognition.py, in line 20.
8. Add your video path to /path/to/vehicle-registration-plate-recognition/src/camera/src/camera/camera.cpp, line 13.
9. Change your directory to /path/to/vehicle-registration-plate-recognition and then enter:

catkin_make

It enables us to use different programming languages(In addition to other benefits). For example, to implement low-level image processing methods, we can use faster languages like C++ compared to Python.
Here we can see the relations graph between nodes in our system.

Open two terminal tabs(or use tmux 😜).

In the first tab:

./launch.sh

and just wait till all initializations are over(takes 20 or so seconds).

Then in the second tab enter:

./start_cycle.sh

Here you go!