针对禁令交通标志牌提出了一种基于HOG-LBP自适应融合特征的交通标志检测方法,将标志图片等分为多个不重叠的块,每块内将加权后的HOG和LBP特征进行串行融合作为最终特征,其中每类特征权值由块内梯度幅值决定,融合后的特征采用SVM进行分类器训练,并将训练结果用于交通标志检测。实验结果表明,基于HOG-LBP自适应融合特征的效果优于基于单独HOG、LBP特征和简单HOG-LBP融合特征的效果。
This paper presents a novel approach of detecting forbidden traffic signs. In this approach, a HOG-LBP (histograms of oriented gradients-local binary patterns) adaptable fused feature is proposed. The traffic sign image is cut into several non-overlapping blocks, in each block, the HOG and LBP features are weighted serial fused based on each block's gradient value. Then to get classifier which is used in detecting forbidden traffic signs, SVM (support vector machine) is used in training features. The experimental resu proposed fusion feature is more effective and feasible than HOG, LBP and simp HOG-LBP feature.