中文摘要：

提出一种基于颜色不变量和塔式梯度方向直方图PHOG（Pyramid Histogram of Oriented Gradients）特征的交通标志检测方法。该方法首先在高斯颜色模型下提取颜色不变量特征并对其进行聚类,以分割出候选感兴趣区域;然后提取感兴趣区域的PHOG特征并用支持向量机进行形状分类,进而区分交通标志形状和噪声区域。对自然环境下的交通标志,PHOG特征所采用的Canny算法在获取感兴趣区域轮廓时会产生较多噪声,从而降低交通标志分类性能。为此,提出利用Chromatic-edge来增强目标轮廓并抑制噪声以提升PHOG特征描述能力。实验结果表明该方法对光照、阴影、遮挡、以及背景复杂等因素具有较高的鲁棒性,获得了较高的检测率和较低的误检率。

英文摘要：

We propose a new detection approach for traffic signs which is based on colour invariants and pyramid histogram of oriented gradients（PHOG） features. The method first extracts colour invariants feature in Gaussian colour model and clusters these colour invariants in order to segment the candidate regions of interests（ROIs）; then it extracts the PHOG features of ROIs and uses support vector machine to classify the shape,and then differentiates the shape and noise regions of the traffic signs. For the traffic signs in natural environment,the Canny algorithm used by PHOG feature will generate more noises when obtaining the contour of ROIs,so that the classification performance of traffic signs is reduced. Therefore,we propose to use Chromatic-edge to enhance the object contour while suppressing the noises so as to boost the descriptive power of PHOG feature. Experimental results demonstrate that our method is quite robust on the factors of illumination,shadow,occlusion and background complexity,etc.,and achieves higher detection rate and lower false positive rate.