影像辨識在全自動化系統之應用研究

The Application Research of Image Recognition on Total Automation System

蘇仲洋(Chung-Yang Su)

指導教授 ： 王永鐘

國立臺北科技大學/電資學院/電機工程系研究所/碩士(2008年)

https://doi.org/10.6841/NTUT.2008.00442

摘要

本論文研發用於停車場進出車道的「車牌辨識系統」，並利用「全自動化系統」提供之「應用程式界面」(Application Interface)進行系統整合。取像工具採用「全自動化系統」既有網路攝影機(IP camera)讀取車輛影像後，以灰階圖、二元圖及邊緣圖轉換技術對車輛影像進行前處理，再利用邊緣圖特性分析取出車牌區域及字元。
車牌字元辨識部份，將正規化為10×6的二元圖點陣圖字元轉換成60×1的輸入陣列，透過60組輸入層神經元、60組隱藏層神經元及36組輸出層神經元之「倒傳遞類神經網路」，來辨認0至9及A至Z的字元。再以VB撰寫整合界面程式，將車牌字元辨識結果寫入「全自動化系統」資料庫，以實現不同系統間之整合應用。
類神經網路的訓練，採用180組不同於辨識樣本的正規化字元來進行批次(batch)訓練，「車牌辨識系統」實驗環境在車輛影像面測量的照度為2600 勒克斯(lux)，並以事先取得九十台車輛影像圖片來進行模擬辨識。結果車牌區成功切割率達90%，每一字元成功切割後的平均辨識率達78.5%，整合界面程式資料傳遞車牌字元辨識延遲小於1秒鐘。

關鍵字

影像處理 ； 車牌辨識 ； 全自動化系統 ； 中央監控系統 ； 類神經網路

並列摘要

This paper is used for research and development of the car park exit/entrance "Car plate recognition system" and use "Total automation system" to provide "Application interface" for system integration. As a tool for "Total automation system" existing network cameras to read images of vehicles, a gray-level, threshold and the edge line conversion technology is for vehicles image pre-processing, and use of the edge line information analysis for plate and characters abstraction.
For car plate character recognition, 10 × 6 of the bitmap characters will be normalized into the importation of 60 × 1 array, through the importation of 60 input layer neurons, 60 hidden layer neurons and 36 output layer neurons of the "back-propagation neural network", to identify the 0-9 and A-Z characters. Furthermore, VB is used to integrate the application program, which can finalize the recognize results of car plate characters to be written into the "Total automation system" database to achieve integration among the different systems.
Neural network training use a group of 180 training samples that identified from the testing characters to carry out batch training-that "Car plate recognition system" in the environment with 2600 lux lightness on the images of vehicles in advance of to obtain prior 90 vehicles images to carry out simulated images recognition. Success plate recognize accuracy is up to 90 percent, after success cutting of each character and the average characters recognize accuracy is up to 78.5 percent; Application interface in car plate character recognition data transmission delay is less than 1 second.

並列關鍵字

Image processing ； Car plate recognition ； Total automation ； Building Manage System ； Neural Network

參考文獻

[8] 劉國偉，移動中車牌定位系統之設計與實作，碩士論文，國立臺北科技大學電機工程系研究所，民國九十四年六月。

[1] Brochure of Honeywell international Inc, EBI overview, SL-54-1532 June 2005.

[2] R. C. Gonzalez and R. E. Woods, Digital image processing, New Jersey: Prentice Hall, 2002.

[5] C. C. Yang and J. J. Rodriguez, "Efficient luminance and saturation processing techniques for bypassing color coordinate transformation", Proceedings of IEEE International Conference on System, Man and Cybernetics, vol. 1, 1995, pp. 667-672.

[7] W. K. Pratt, Digital image processing, New York: John Willy & Sons, 2001.