一种非监督道路场景分割方法

针对道路场景分割中训练样本量大、不同类型道路过渡中易产生误分割的问题,该文提出了一种非监督的道路场景分割方法。首先用K均值聚类对第一幅图像进行初始化分割,再用图割法对其进行能量最小化的优化分割,最后用优化后的分割图像重新计算类别中心,用于指导下一帧图像的图割优化分割。实验表明,该方法无需大量训练样本,可以快速地对道路场景进行分割,还能够在不同的道路类型过渡过程中保持很好的分割效果。

Unsupervised Road Scene Segmentation Method

To solve the problems that lots of training samples are needed in the road scene segmentation and the changes of different roads cause the segmentation error easily,this paper proposes an unsupervised road scene segmentation method.First,K-means clustering method is applied to the first image for its initial segmentation;Second,graph cut optimization algorithm is used to minimize the total image energy to get the optimal segmentation.With the computed class centers of the segmented image,the next image is also optimized by graph cut.Experimental results show that this method can segment the road scene quickly without quantities of training samples,and can keep efficient in changing of different road types.