戶外交通號誌辨識之研究

Abstract

在此論文中，一張影像中交通號誌和其他物體之間，使用HSI彩色座標系統和飽和度加強得到較好的分離效果。傳統的圖樣辨識方法只有抽取那些因位移、旋轉、大小改變而沒有改變的特徵向量，但是這些特徵卻沒有因扭曲變形或遮蔽等變化而沒有改變，所以這篇論文描述一個交通號誌辨識系統的方法可以容忍以上的變化，特徵抽取的方法是基於傅立葉轉換、極座標轉換、和離散cosine轉換(DCT)的技術，由於DCT的能量緻密性，DCT可以用來當形狀辨識的特徵抽取。這些特徵被用來當一個分類器神經網路的輸入。最後利用外在環境中所可能發生的各種情況來評價我們所提出的系統的表現。

In this thesis, the hue, saturation, intensity (HSI) coordinate system and saturation enhancement are used to get the better segmentation between the traffic sign and other objects in the image. The conventional pattern recognition methods only extract the feature vectors which are invariant to translation, rotation, and scale but are not invariant to distortion or occlusion in outdoor environment. This thesis describes a traffic sign recognition system capable of tolerating the above variations. The feature extraction is based on the techniques of the Fourier transform, Log-polar transform, and the of the discrete cosine transform (DCT). Because of the energy compaction of DCT, the DCT can be used as feature extraction for shape recognition. The features are used as the inputs of the neural network which is a classifier. The performance of the proposed system is evaluated by examining the effect of various conditions which may occur in natural outdoor environment.