應用兩階段分類法於交通標誌偵測與辨識之研究

Applying the Two-Stage Classification Method for Road Sign Detection and Recognition

林建仲(Chien-Chung Lin)

指導教授 ： 郭文嘉

元智大學/資訊學院/資訊管理學系/碩士(2006年)

https://doi.org/10.6838/YZU.2006.00148

摘要

本論文提出一個應用兩階段分類法於交通標誌偵測與辨識的方法。交通標誌之自動化偵測與辨識除了可以輔助駕駛自動化外，更可以提供智慧型運輸系統應有的即時訊息，以提高自動駕駛的安全性與可信賴度。除此之外，亦可輔助駕駛者注意行車安全，保障駕駛者與行人之安全。本論文主要分偵測階段與辨識階段兩部分：偵測交通標誌階段，透過標誌之幾何特性，運用霍氏轉換之計算、角點之偵測以及投影之方式，找出交通標誌之影像位置，並使用三角形與圓形之幾何特徵去除不相關之背景。辨識階段則利用迴旋積及半徑式函數類神經網路與K-d樹搜尋辨識兩階段做分類辨識分析：首先利用半徑式函數類神經網路進行分群，以期降低辨識錯誤之比率。最後針對半徑式函數類神經網路所區分之每一群，各自利用特徵值切割建立一個K-d樹來進行搜尋辨識，K-d樹除了可以搜尋辨識外，亦可以修正在半徑式函數類神經網路中分群錯誤之圖像。實驗結果顯示，大部分的交通標誌均可以成功的偵測與辨識，辨識率可到達95.5%，同時對於特殊狀況影響之標誌亦均能有效地偵測與辨識。本論文所提之方法將有助於日後智慧型輔助運輸系統之發展，進而提供即時有效之道路標誌輔助訊息。

關鍵字

霍氏轉換 ； 投影 ； 幾何特性 ； 迴旋積 ； 半徑式函數類神經網路 ； K-d樹

並列摘要

This study propose a road sign detection and recognition method using two-stage classification method. The automation of road traffic sign detection and recognition not only can afford the information immediately to the intelligent transportation system, but also increase the security and the reliability of automatic driving. Besides, it can assist the driver to pay much attention the safety of driving. In the detection stage, geometric characters of road traffic signs, hough transform, corner detection, and projection are used to detect the exact position of the road traffic sign in the image. Also, the properties of triangle and circle are adopted to eliminate the irrelative background region. In the recognition stage, convolution, radial basis function neural network and K-d tree are used to recognize the road signs. We use the radial basis function neural network to group the possible candidates to decrease the error rate of false recognition. The K-d trees are then constructed with separate features according to the classification results of radial basis function neural network in the previous step. The K-d tree can also be used to modify the results of classifying the road sign into the wrong group. Experimental results show that most road signs can be correctly detected and recognized by our proposed method with the accuracy of 95.5%. Moreover, the method is robust against the major difficulties of road sign detection and recognition. The proposed approach would be helpful for the development of intelligent Driver Support System and to provide effective driving assistance message.

並列關鍵字

Hough transform ； Projection ； Geometric characters ； Convolution ； Radial basis function neural network ； K-d tree

參考文獻

[1] W. G. Shadeed, D. I. Abu-Al-Nadi, M. J. Mismar, “ROAD TRAFFIC SIGN DETECTION IN COLOR IMAGES,” in Proc. 10th IEEE Int. Conf. Electronics, Circuits and Systems, ICECS 2003, 14-17 Dec., 2003, Sharjah, United Arab Emirates, vol. 2, pp. 890–893.

[2] M. Bénallal, J. Meunier, “Real-time color segmentation of road signs,” in Proc. IEEE Canadian Conf. on Electrical and Computer Engineering, IEEE CCECE 2003, 4-7 May, 2003, Canadian, vol. 3, pp.1823–1826.

[3] S. H. Hsu, C. L. Huang, “Road sign detection and recognition using matching pursuit method,” Image and Vision Computing, vol. 19, no. 3, pp. 119-129. Feb. 2001.

[5] A. de la Escalera, L. E. Moreno, M. A. Salichs, and José María Armingol, “Road Traffic Sign Detection and Classification,” IEEE Trans. Industrial Electronics, vol. 44, issue 6, pp.848-859, Dec. 1997.

[6] T. Askaura, Y. Aoyagi, and K. Hirose, “Real-Time Recognition of Road Traffic Sign in Moving Scene Image Using New Image Filter,” in Proc. 39th SICE Annual Conference, SICE 2000, Int. Session Papers , 26-28 July, 2000, pp.13-18.