基于MSER和SVM的快速交通标志检测

为解决传统的基于机器学习的交通标志检测(TSD) 方法需要对每一个待检测子窗口进行处理而导致算法实时性不高的问 题,提出了一种基于感兴趣区域(ROI)提取和机器学习的快速TSD 算法。针对传统基于颜色阈值的ROI提取方法具 有对光照变化较敏感等缺点,设计了一种颜色增强下的最大稳定极值区域(MSER)方法 ,根据标志的颜色进行 颜色增强,对颜色增强图像提取MSER得到交通标志ROI；然后在图像的多尺度滑动遍历检测 过程中,仅对包含ROI的滑 动窗口进行方向梯度直方图(HOG)特征的提取,并通过支持向量机(SVM)进 行分类判别。实验结果表明,本文改进的TSD方法在运算速度上有较大提升,具有很好的鲁 棒性,且获得了96.42%的检测率以及较低的误检数。

Rapid traffic sign detection based on MSER and SVM

In order to solve the problem that the real time performance is not hi gh since processing needs to be carried out in each detected sub-window when traditional methods based on machine learning are used in traffic sign detection,a rapid traffic sign detection (TSD) algorithm based on region of interest (ROI) extraction and machine learning is proposed in this paper.Aiming at the shortcomings that traditional ROI extraction method based on color threshold has more sensitive to illumination changes,we design an effective method based on maxima lly stable extremal regions (MSER) under color enhancement.In this method,image is firstly enhanced by the color characteristics of traffic signs,and the ROI of traffic signs is obtained by extracting MSERs from co lor enhanced image.Then in the process of multi-scale sliding and traversing window detection,histogram of oriented g radients (HOG) feature is extracted only in the sliding window which contains the ROI of traffic sig ns.And finally,classification and discrimination are completed through the support vector machine (SVM).Experimental results show that the computing speed of presented approach in this paper has improved greatly.It also has good robustness,and can achieve accuracy 96.42% of and a low false alarm number.