基于先验信息的交通信号灯识别方法

A Traffic Light Recognition Method Based on Prior Information

针对基于颜色的交通信号灯识别(TLR)准确率不高的问题,文章提出一种融合色调、饱和度、明度(HSV)颜色空间和图像灰度信息的交通信号灯识别方法。先将图像进行倾斜校正和大小标准化,以满足融合交通信号灯识别方法的需求。然后,将红绿蓝(RGB)图像转化为色调、饱和度、明度颜色空间,对交通信号灯进行初步识别。通过霍夫直线检测提取交通信号灯的矩形边界,实现感兴趣区域(ROI)检测。针对色调、饱和度、明度颜色空间无法识别的交通信号灯图像,以交通信号灯的结构特点做为先验信息,进行灰度图像感兴趣区域像素值求和,进一步识别交通信号灯。通过实验对该融合交通信号灯识别方法进行了分析,并与单独使用色调、饱和度、明度颜色空间和灰度图像感兴趣区域像素值求和的方法进行对比,验证文章算法的有效性。

To solve the problem of low accuracy of color-based traffic light recognition(TLR),a TLR method that combines HSV(Hue,Saturation,Value)color space and gray image information is proposed.First,image tilt correction and size standardization is needed to meet the needs of the proposed fusion traffic light recognition method.Then,the RGB(Red,Green,Blue)images are converted into HSV(Hue,Saturation,Value)color space to initially recognize the state of traffic light.The boundaries of traffic lights is extracted by hough line detection,and region of interest(ROI)detection is realized.Considering that some traffic light images cannot be recognized by HSV(Hue,Saturation,Value)color space method,and the structure characteristics of the traffic light are used as the prior information,and the region of interest pixel values of the gray image is summed to further identify the state of traffic light.The fusion traffic light recognition method is evaluated through experiments and compared with the methods that separately use HSV(Hue,Saturation,Value)color space and gray image region of interest pixel values summation to evaluate the performance of the proposed algorithm.