中文摘要：

本文提出一种基于双目立体视觉的场景分割方法：首先根据双目立体视觉系统提供的左右视图进行三维场景重构,得到场景的几何深度图,同时利用左视图进行RGB颜色空间到CIELab均匀颜色空间的转换以得到颜色信息；然后将颜色与几何信息构造生成六维向量；最后再将六维向量给到聚类算法中进行分割并对分割的伪影进行消除,得到最终的分割结果。对MIDDLEBURY数据集样本场景baby 2实验了六种立体视觉算法和三种聚类技术的不同组合进行的场景分割,从实验结果来看,不同的组合应用本文所提方法都比传统方法具有更好的分割效果。

英文摘要：

A scene segmentation approach based on binocular stereo vision is proposed. Firstly, an 3D scene is reconstructed based on the left and right view of binocular stereo vision system, and then the scene geometry depth maps were obtained. Meanwhile, RGB color space of the image from left view is converted to CIELAB uniform color space to obtain color information. After that, a 6D vector is constructed by both color and geometry information. Finally, the 6D vector is given to clustering algorithm to segment the scene and remove the artifacts, and at last the final segmentation results are obtained. The Middlebury data set sample scene baby 2 have been segmented with different combinations of stereo vision and clustering techniques. Experimental results show that the proposed method can obtain a better segmentation than the methods based on just color or just geometry.