基于先验信息的交通标志检测

Traffic Sign Detection Based on Prior Information

道路交通标志的自动识别通常应用于车辆的自动或半自动辅助驾驶系统中,为车辆驾驶提供对周围环境的理解。然而现有的交通标志检测与识别算法针对对象比较单一,多以某一类中的若干个标志为检测对象,当检测对象的样本数较多时,检测正确率明显下降。此外,该类方法虽然考虑了交通标志的颜色和形状等信息,但却忽略了颜色、几何形状与标志之间的确定性关系。提出了一种快速有效的交通标志检测算法,根据标志的形状及颜色等先验信息,建立一棵交通标志决策树,逐层筛选兴趣区域,并根据交通标志的轮廓信息将交通标志检测结果分为十个子类,通过子类结果及交通标志的先验信息进行交通标志检测。实验结果表明,当交通标志被遮挡时,该方法降低了交通标志检测的漏检率以及误检率。所提出的方法降低了TSR(Traffic Sign Recognition)系统的复杂性,提高了系统的实时性和鲁棒性。

The automatic recognition of traffic signs can be applied to the automatic or semi-automatic auxiliary driving system to provide the information of surrounding road conditions.However the existing algorithms are relatively unitary for detection object and mostly only detect several signs in a certain class.With the increasing of traffic signs,the correct rate of detection is decreased obviously.In addition,these methods consider the color and shape,but ignore the deterministic relation between shape and geometric with traffic signs.A fast and efficient algorithm of traffic sign detection is proposed.According to the prior information of the shape and color of traffic signs,it establishes a decision tree of traffic signs,which can filter out interesting regions and divide traffic signs into 10 sub classes,detecting traffic signs by sub-classes results and the prior information of traffic sign.The experimental results showthat when the traffic signs obscured,this method reduces the residual rate and false detection rate in traffic sign detection.The algorithm can reduce the complexity and improve the real-time performance and efficiency of the TSR system.