Licence Plate Detection and Recognition

The project implements an image-based Deep Learning pipeline to detect license plates and read the registration number. The algorithm is based on a 2 steps inference model:

• Mobilenet SSD-Lite to detect License plates within greyscale 320x240 images
• LPRNet to read the regristration number from 94x24 Licese Plates crops

Models are invoked in sequence to run the full pipleine.

Getting Started

To generate the code for the two models:

make -f ssd.mk clean model
make -f lprnet.mk clean model

After both models are generated, to run the full-application:

make clean all run

AT Optimization

The utils/call_at.py script launches the Autotiler engine with different L2 memory configurations and plots the trend of several metrics to explore the optimal configuration:

python3 utils/call_at.py SSD

and/or

python3 utils/call_at.py LPR 

Test nntool models

Once you generated the models you can test the results in NNtool.

For the LPRnet:

nntool BUILD_MODEL_LPR/lprnet.json
set input_norm_func 'x:x/128-1' [there is not a formatter so you need to specify a normalization]
dump images/china_1_cropped.ppm -q -S nntool --mode RGB -T
run_pyscript utils/get_char_from_nntool.py [to see the predicted characters]

For the SSD model:

nntool BUILD_MODEL_LPR/lprnet.json
set input_norm_func 'x:x/128-1' [there is not a formatter so you need to specify a normalization]
dump images/china_1.ppm -q -S nntool --mode RGB -T -v

NOTE:

• -q: quantized inference like GAP
• --mode RGB: convert to RGB (model wants 3 channels images)
• -T: transpose the RGB images to meet the CHW execution order of nntool/Autotiler
• -v: allows the user to visualize the detected bounding boxes

Test tflite models

The utils folder contains two Python scripts (lprnet_tflite_inference.py and ssd_tflite_inference.py) to test the tflite quantized models:

python3 lprnet_tflite_inference.py [path/to/image]

if not specified the test images will be taken from the images folder.

Debug

To test the two models separately:

make -f ssd.mk clean all run
make -f lprnet.mk clean all run

The result from the ssd model can be copied and pasted in the utils/DrawBB.py script to see the detected bounding box.

Both models can be run in __EMUL__ mode for debugging purpose with:

make -f emul.mk clean all MODEL=1
./lprnet_emul images/image.ppm > log.txt

or

make -f emul.mk clean all MODEL=2
./ssdlite_ocr_emul images/image.ppm > log.txt