發展自動車牌辨識之機器學習系統

Development of a Machine Learning System for Automatic License Plate Recognition

范哲綸(Che-Lun Fan)

指導教授 ： 陳冠宇

中原大學/工學院/機械工程研究所/碩士(2017年)

https://doi.org/10.6840/cycu201700860

摘要

隨著科技的進步，自動車牌辨識系統的應用已愈趨普及。一般來說，車牌辨識可概分成四個階段：影像前處理、影像校正、影像分割及字元辨識。本文的研究目的即著重在整合機器視覺及機器學習二個領域，以發展基於類神經網路之機器學習系統應用於自動車牌辨識之字元辨識上，提高其辨識效能。首先，本文發展一套圖形化使用者介面以利車牌辨識系統之運作，接著，架設一部攝影機拍攝汽車之車頭，擷取的影像需進行色彩空間轉換、灰階閥值篩選、影像二值化、邊緣偵測、影像形態處理…等步驟後，再從中取得類神經網路之訓練樣本，經由類神經網路的反覆循環訓練學習後，最後進行系統實測。本文共測試96組汽車影像，獲致約92%的辨識成功率，其中辨識失敗原因有車牌固定螺絲鏽蝕痕跡與字元連結、影像失焦模糊、影像拍攝角度過於偏斜…等，整體而言，若原始影像品質許可，本文發展之系統的辨識成功率還可提高。

關鍵字

機器學習 ； 車牌辨識 ； 影像處理

並列摘要

With the progress of science and technology, the automatic license plate recognition (ALPR) system and its applications have increased noticeably. In general, the ALPR system can be divided into four major stages. These stages are image pre-processing, image calibration, image segmentation and character recognition. The purpose of this thesis is to develop an ALPR system based on neural networks, especially used for enhancing performance of character recognition, by integration of machine vision and machine learning. First, this thesis develops a graphical user interface to facilitate the operation of the ALPR system. Next, car front images can be grabbed by a camera and need to be performed image pre-processing, such as color space transformations, gray scale thresholding, image binarization, edge detection, morphological image processing. Accordingly, we can obtain training samples for used in the learning phase of the ALPR system. Finally, physical testing of the ALPR system in real time is performed to evaluate the performance after the training of the neural network is completed. In this thesis, we tested 96 car front images and the ALPR system achieved a 92% success rate. Some common causes of failure in recognition include characters corrupted by the rust from license plate fixed screws, blurred or unfocused images, and the shooting angle was too skewed. As a whole, if the original quality of grabbed images are good enough, the success rate of the proposed ALPR system can be improved.

並列關鍵字

machine learning ； license plate recognition ； image processing

參考文獻

[6] 何宇翔，利用階層式類神經網路實現車牌辨識系統，嶺東科技大學資訊科技應用研究所碩士論文，2000。

[17] 蘇仲洋，影像辨識在全自動化系統之應用研究，國立台北科技大學電機與控制工程學系碩士論文，1998。

[5] 張簡子，以小腦模型在 FPGA 上作車牌辨識，國立台灣師範大學工業教育所碩士論文，1990。

[7] 魏銪志，動態多標的車牌辨識系統之研究，元智大學資訊研究所碩士論文，2000。

[11] R. C. Gonzalez and R. E. Woods, Digital Image Processing, New Jersey: Prentice Hall, 2002.