基于局部线段模式特征的城市道路视觉检测

Urban road vision detection based on local line pattern feature

局部二值模式(Local Binary Pattern,LBP)特征对于局部纹理信息的提取非常精细,但对于发散的远端信息则会遗漏部分关键特征而造成检测精度降低。针对这一问题,文中提出了局部线段模式(Local Line Pattern,LLP)特征算法。首先构建城市道路数据库并将道路图像进行分块处理,然后用LLP算子提取直方图来描述局部纹理结构,并将其串联得到特征向量,最后结合BP神经网络实现城市道路可行驶路面检测。实验结果表明,LLP特征算法在检测精度上优于现今主流的各类改进型LBP算法。LLP结合"等价模式"LBP算法的实验结果显著优于"等价模式"LBP算法,在城市道路数据库中检测精度较"等价模式"LBP算法提高了4. 3%。

Local binary pattern(LBP)feature has the advantage of finely extracting local texture information.However,for missing divergence far-end information,missing key features result in low urban road detection accuracy.To solve this problem,local line pattern(LLP)feature algorithm is proposed.Firstly,the urban road database is constructed and the road image is processed in blocks.Then the histogram is extracted by the LLP operator to describe the local texture structure,then it is connected to form a series feature vector.Finally,by combining LLP feature algorithm with the BP neural network,the urban road surface detection can be achieved.Experimental results show that the LLP feature algorithm is superior to the current main-stream LBP algorithms in detection accuracy;LLP combined with uniform pattern LBP algorithm is more accurate than that of the uniform pattern LBP algorithm.Its detection accuracy in the urban road database is improved by 4.3%compared with the“Uniform Pattern”LBP algorithm.