[M480] Car License Plate Recognition Based on Keras

[chhsieh3]

• Application: This document describes how to program Keras weight for deep learning technology to develop car plate recognition, and to help users to implement the car plate recognition on the NuMicro® M480 series microcontroller.
• BSP Version: M480 BSP CMSIS V3.05.001
• Hardware: NuMaker-IoT-M487 Ver1.2

The global market of machine learning is flourishing recently with the advancement of technology. Machine learning refers to an accumulative and autonomous behavior enhancement from a machine through a series of learning process. The learning process is to give training data to mathematical data models, and it could be categorized into supervised, unsupervised and reinforcement learning. The idea of machine learning can be realized in almost every field; social media features, product recommendations on the Internet, image recognition, and language translation are all examples of machine learning.

With DNN (Deep Neural Networks) and CNN (Convolution Neural Networks) that support machine learning networks, the NuMicro® M487 Ethernet series from Nuvoton, a high performance and low power microcontroller, is suitable to be used in related applications. A M480 platform (with M487 on it) could recognize any vehicle license plate by using its learning neural network algorithms. A CMOS sensor is required to capture the plate image and it takes around 200 ms to identify the image. The M487 supports the image resolution of QVGA 320 x 240.

This sample code captures car license plate, using machine learning neural network algorithms for identification. The Arm® Cortex®-M4 core supports the DSP instruction to speed up algorithms. The information can be transmitted to a peripheral such as UART on edge device.

You can download the sample code at https://www.nuvoton.com/resource-downlo ... 0118165849