基于决策树型SVM的交通标志图像识别

由于采集信息装置简单和外界环境复杂，以及对识别方法的实时性、准确性要求，使得交通标志识别成为一项难题．首先根据交通标志特殊颜色信息和规则几何外形，利用边界矩技术迅速清除干扰区域，然后将剩下的区域规格化，送入训练好的决策树型支持向量机识别．在决策树型向量机训练阶段，使用模糊聚类算法，较好地完成了树型建构，使向量机具有良好的区分度．对大量实景图像进行实验证明，本研究方法具有平移、旋转、缩放、拉伸不变性和较强的容噪能力．

Recognition of Traffic Sign Images Based on Decision-tree-based Support Vector Machine

Because of the simplicity of information acquisition device, the complexity of outer environment and the required real time and accuracy of recognition method, the recognition of traffic sign becomes a challenging task. At first, the authors quickly clear the disturbing regions using shape based moments on the basis of particular color and regular geometrics of traffic signs, then make remaining regions standard and input them into decision-tree-based support vector machine which well worked to recognize. Fuzzy clustering algorithm is used during the process and it performs the structure of decision tree and provides support vector machine good classification ability. Using many real road images, experimental results prove that the method is simple and efficient, having affine invariability of translation, rotation, scale and distortion and being noise-proof.