Description

Just a DEMO to play around. Won't work on every mobile. This version has lesser resolution etc. for speed. For the encoder model NN API delegate is deactivated for all devices except Google, as errors and not working. Both models optimized for speed and size.

Features

- No internet used
- 1-liner
- Image Recognition with TensorFlow 2.x based on LPRNet
- TensorFlow Object Detection API with TensorFlow 2.x

DEMO Features

- GDPR compliant
- No text selection or copy
- No pictures are saved or uploaded
- No internet is used or any data is resolved
- No logging
- Process image size: 1024x768
- Detection image size: 640x640
- Maximum 8 license plates shown at once
- NN API delegate is deactivated because of issues on Samsung/Huawei devices, maybe will affect performance.
- Most country patterns will fail or show wrong results, I just fixed some EU country patterns
- Both models minimal optimized for speed and size
- Allowed the just image recognition mode in the DEMO (slide the switch to left)

Tested

Please bare with me and keep in mind, it won't work on every mobile, it is a specific DEMO, and I haven't checked all specific hardware vendor differences... It is as well not battery optimized, as I am using minimum 4 threads, it will surely drain your battery, once you start processing the images.

Versions

Only Android versions starting from 8.1 to 11 were tested.

Hardware

- Google Pixel 5 (Android 11) - deactivated NN API delegate for the encoder
- Google Pixel 2 XL (Android 10/11)
- Google Nexus 6p (Android 8.1)
- Samsung Galaxy S8/S9/S10 - deactivated NN API delegate for the encoder
- Unspecific Huawei - deactivated NN API delegate for the encoder

Links and sources

- My "quick" journey creating license plate recognition on Android with TensorFlow 2
On mobile HTML-links are not shown:
https://medium.com/swlh/my-quick-journey-creating-license-plate-recognition-on-android-with-tensorflow-2-51e89387fac4
- LPRNet: License Plate Recognition via Deep Neural Networks (Sergey Zherzdev, Alexey Gruzdev)
- Photo credits by Roman Kraft
- TensorFlow* Object Detection API