Author(s)

Wang Yibo, Zhou Yujun

Corresponding Author

Wang Yibo

Abstract

In this paper, a car license plate recognition system based on dual-core DSP OMPL138, is proposed. The core algorithm consists of four parts: image pre-processing, license plate localization, character segmentation, and character recognition. An adaptive filtering method is used to solve some image blurring and unclear problems caused by external environment. The Roberts operator is used to edge detection; A minimum threshold selection optimization genetic algorithm is used to get a complete character segmentation; Finally, a neural networks algorithm based on three different training models, is used to achieve license plate character recognition. Such program is developed in a cross-compiling Qt platform and runs in the ARM-Linux environment on the TL138F-TEB experiment box, in which the DSP core executes the main calculation tasks while the ARM core outputs the results of recognition, and these dual cores make data exchange through TI’s inter-core communication framework (syslink). The final test results show the differences in the accuracy of license plate recognition achieved by three different training models.

Keywords

Car license plate recognition, neural networks, Cainny edge detection operator