基于曲率尺度空间角点检测的交通标志分离算法

Separation Algorithm of Traffic Signs Based on Curvature Scale Space Corner Detection

针对交通标志出现互连现象导致检测率下降的问题,提出了一种基于曲率尺度空间(CSS)角点检测的交通标志分离算法。使用基于红绿蓝(RGB)归一化的彩色分割算法和区域特征判决准则自动识别多标志互连候选区域,并对提取的目标区域进行边缘平滑和轮廓跟踪。利用基于全局和局部曲率特性的CSS角点检测器对提取的轮廓进行角点检测,并依据角点凸凹性判定准则及分离点对匹配条件,从角点中提取标志间的分离点对。利用Bresenham算法寻求分离点对间的分离线,实现标志的最终分离。实验结果验证了算法的有效性,与现有基于分水岭变换的标志分离算法以及改进的自适应分离算法相比,克服了标志过度分离问题,提高了标志检测整体性能。

In view of the detection rate degradation of traffic sign due to multiple signs interconnection, a separation algorithm based on curvature scale space （CSS） corner detection is proposed. The candidate regions of multiple signs interconnection are automatically identified based on red green blue （RGB） normalization color segmentation algorithm and region feature decision criterion. The edge smoothing and contour tracking for the extracted target region are performed. For the obtained contour, the corner detection by using the CSS corner detector based on global and local curvature properties is conducted. And according to the criterion for judging the convexity- concavity of corner and separation points pair matching condition, the separation points pair between the interconnection traffic signs is extracted from the detected corners. The separation line between the separation points pair is achieved by adopting Bresenham algorithm, thus the traffic signs are separated by using the line. Experimental results verify the effectiveness of our method. The proposed method overcomes the problem of the traffic signs over-separation and improves the overall detection performance compared with the existing separation algorithm based on watershed transform and the improved adaptive separation algorithm.