基于支持向量机的实时路面检测算法

Real-time Road Detection Algorithm Based on Support Vector Machines

二阶多项式核函数支持向量机分类决策函数可以表示为待分类向量各分量的形式,其中的同类项可以合并,同类项的系数在得到支持向量后可以计算得出。使用这样的分类决策函数,可以避免分类时待分类向量和各个支持向量逐个进行的运算,使分类计算速度和支持向量个数无关。针对实际道路图像的对比实验表明,采用这种新算法的支持向量机路面检测分类器,在路面检测精度上优于神经网络,在计算速度上也能很好地满足实时检测的要求。

For a two order polynomial kernel function based support vector classifier, the classifying function can be written in the form of the vector＇s components with the similar terms combined. When the support vectors are got, the numerical coefficient of those terms can be calculated. Classifying an unknown class vector by such a classifying function, the calculation between each support vector and the vector to be classified can be avoided, which means that the speed of classification is independent of the number of support vectors. Experiments with real road images show that such a support vector classifier is superior to the neural nets in preciseness for road detection, and the classifying speed can meet the real-time computing need well.