Segmentation-based classification for 3D point clouds in the road environment

Classification of 3D point cloud in urban scenes has been widely applied in the fields of automatic driving, map updating, and change detection. To develop the accurate and effective method for the classification remains a challenge. We propose a three-stage framework to classify 3D point clouds acquired in the urban road environment. First, an efficient segmentation approach is applied to generate segments as the entity to be classified. This is achieved by using the pairwise linkage (P-Linkage) algorithm for the initial point cloud segmentation, followed by two-step post-processing approach to improve the segmentation results. Second, a set of novel features is extracted from each segment and its performance is evaluated using four classifiers: Support Vector Machine, Random Forests, Naive Bayesian, and Extreme Learning Machine. Third, the contextual constraints among objects are used to refine the classification via graph cuts. The results showed that the initial classification can obtain a high average precision of 80.8-92.9% and a good average recall rate of 77.5-79.1%. After refinement via graph cuts, the precision and recall rate increase by approximately 0.3% and 3.1%, respectively. Our framework is effective for classifying 3D urban point clouds acquired by a mobile laser scanning (MLS) system in the road environment.