点击次数：

论文类型：会议论文

发表时间：2016-07-27

收录刊物：EI、CPCI-S、Scopus

卷号：2016-August

页面范围：7131-7136

关键字：Fast Scene Segmentation; 3D Point Clouds Classification; Conditional Random Field; Urban Environments

摘要：3D scene segmentation and classification are two challenging tasks for unmanned ground vehicles (UGVs) working in urban environments. According to the structured characters existing in urban scenes, a novel 3D scene segmentation approach is proposed based on region growing algorithm using sequential 2D laser scanning data. Since no iteration procedure is used in this approach, it can perform fast segmentation than traditional ones such as k-means based segmentation approach. A Conditional Random Field (CRF) model is utilized in our urban scene classification framework, in which each segmented patch is selected as a CRF node and the high-order cliques are generated by a clustering algorithm. The experimental results based on DUT2 dataset and KAIST dataset are given to demonstrate the validity and robustness of the proposed approach.