Abstract: Traffic signs consist of outer shapes and internal designated symbols, and the former can be described by Histogram of Oriented Gradients（HOG）which, however, is influenced by noise, and details of the latter can be well depicted by Local Binary Patterns（LBP）. This paper, based on blocking HOG-LBP fusion features, proposes self-adaptation identification method of traffic signs. Firstly, it through weight coefficient which is attained by calculating blocking HOG, makes a judgement of which part some certain block belongs to（outer shapes incline to adopt HOG, and internal designated symbols, the LBP features）, and secondly, it puts features created by mixing self-adaptation serials into support vector machine to conduct recognition. Simulation experiment shows that this algorithm achieves 100% recognition rate of standard traffic signs, and 76% of nonstandard traffic signs including unclear, incomplete and covered ones.

Key words: recognition of traffic signs, self-adaptation fusion features, Histogram of Oriented Gradients（HOG）, Local Binary Patterns（LBP）

摘要： 交通标志由外部轮廓和内部指示符号组成，HOG特征可较好描述图像轮廓但易受噪声影响，而LBP特征对图像细节刻画好，提出基于分块HOG-LBP自适应融合特征的交通标志识别方法。通过分块计算梯度直方图得到的权重系数，来判断该块是属于轮廓还是内部指示，对前者选择HOG权重大，后者选择LBP特征权重大，将自适应串行融合后的特征送入支持向量机识别。仿真实验结果表明，该算法对标准交通标志识别率可达到100%，对含模糊、残缺、遮挡等非标准交通标志也达到了76%。

关键词: 交通标志识别, 自适应融合特征, 方向梯度直方图（HOG）, 局部二值模式（LBP）