基於灰階影像之速限路標偵測與辨識

Speed Sign Detection and Recognition Based on Gray Image

黃雅軒 ； 顏華慶 ； 李允善

《Journal of Information Technology and Applications(資訊科技與應用期刊)》 4卷4期 (2010/12) Pp. 182-192

https://doi.org/10.6302/JITA.201012_4(4).0003

摘要

本論文利用電腦視覺處理來研發一套基於灰階影像的速限路標偵測與辨識技術，進而整合出具有即時處理能力的速限路標偵測與辨識系統，可在不同的路況中提醒駕駛人有關行車速度限制資訊。本技術包含四個主要的處理模組：（1）速限路標偵測、（2）速限路標切割、（3）速限路標辨識和（4）速限路標整合。在速限路標偵測方面，採用了Adaboost演算法，可快速和穩定地偵測出路標的位置和範圍；在速限路標切割方面，利用歐祖（Ostu）二元化、Canny邊緣偵測、橢圓尋找、影像強化和型態學（Morphology）等影像處理的運算，可以有效的切割出路標中數字部份的影像；在速限路標的辨識方面，針對包含已被切割出來數字而形成的矩形區域影像，以梯度直方圖（Histogram of gradient，HOG）進行特徵比對和以原始影像為特徵來做支持向量機（Support Vector Machine，SVM）比對，再將兩種比對方法的結果做線性組合，來產生最終的辨識結果。在速限路標整合方面，使用了速限路標不重複出現和具連續的特性，可大幅降低因背景部份被偵測為路標而造成的錯誤，以及提高辨識的正確性。經由實驗測試，針對在真實路況中所拍攝的758張速限路標樣本，本研究方法可得到高達98.1%的正確辨識結果。整個系統已有效地整合速限路標偵測、速限路標切割和速限路標辨識等處理步驟，完成了一個針對灰階影像既快速且穩定的速限路標偵測和辨識系統。

關鍵字

速限路標切割 ； 梯度直方圖 ； 支持向量機

並列摘要

The study, on the basis of gray image, uses the skill of computer vision processing to develop the technology of speed sign detection and recognition, and furthermore integrates a system of speed sign detection and recognition that has real-time processing ability to remind drivers of speed limitation when stopping vehicles in different road conditions. The technology includes four main processing modules: speed sign detection, speed sign segmentation, speed sign recognition and speed sign integration. In the aspect of speed sign detection, the Adaboost algorithm is adopted to quickly and stably detect the positions and ranges of the signs. In the aspect of speed sign segmentation, the image processing algorithms, such as Ostu algorithm, canny edge detection, ellipse fitting, image enhancement and morphology operation are utilized to efficiently cut out the image in the figure part of sign. In the aspect of speed sign recognition, for the formed rectangular image, including the figures cut out, its characters are validated by Histogram of Gradient (HOG), and the characters of original image are validated by Support Vector Machine (SVM). Afterward, a linear combination is made by the results of the two validation ways to generate the final recognition result. In the aspect of speed sign integration, taking advantage of the characters of no repeated appearance and continuity in speed sign to reduce the errors caused by the background to be detected as a sign and furthermore to enhance the accuracy of recognition. In the aspect of system performance, Support Vector Machine (SVM) and character validation are mixed on recognition results to generate a high recognition performance up to 98.1%. The whole system efficiently integrates those processing steps, such as speed sign detection, speed sign segmentation, and speed sign recognition, and completes a rapid and stable speed sign detection and recognition system for gray image.

並列關鍵字

Adaboost ； Speed Sign Segmentation ； HOG ； SVM