Search

To reduce the cost of manually annotating training data for supervised classifiers, we propose an automated approach to extract training data of urban objects in six classes: buildings, fences, man-made poles, vegetation, vehicles, and low objects. In this study, two segmentation algorithms are firstly implemented to generate meaningful objects from the non-ground point cloud. Then, we generated valid strict rules to label partial RANSAC (Random Sample Consensus) planes and meaningful objects as training data. The strict rules are built upon the semantic knowledge formed by the features of geometric, eigenvalue, RANSAC plane, multidimensional slice, and relative location. The accuracy of strict rule-based (SRB) training data is higher than 98.5 % for buildings, man-made poles, vegetation, and vehicles. The accuracy of low objects and fences reaches 97.10 % and 94.99 %, respectively. Finally, we compared the performance of the KPConv and PointNET++ networks trained by SRB and manually labeled training data to evaluate the effectiveness of our training data. The KPConv overall accuracy using manually labeled and (SRB) training data are 91.5 % and 86.8 % in the Paris dataset, 95.6 % and 92.0 % in the Freiburg dataset, respectively. The experiments demonstrate that automatically labeled training data can achieve similar accuracy compared to manual labels when coupled with two deep learning networks. Therefore, SRB training data extraction can effectively deal with the problem of training data scarcity and provide significant advancements in urban point cloud classification, where manual labeling of training data remains a crucial challenge. [ABSTRACT FROM AUTHOR]

Objects are formed by various structures and such structural information is essential for the identification of objects, especially for street facilities presented by mobile laser scanning (MLS) data with abundant details. However, due to the large volume of data, large variations in point density, noise and complexity of scanned scenes, the achievement of effective decomposition of objects into physical meaningful structures remains a challenge issue. And structural information has been rarely considered to improve the accuracy of distinguishing between objects with global or local similarity, such as traffic signs and traffic lights. Therefore, we propose a structural segmentation and classification method for MLS point clouds that is efficient and robust to variations in point density and complex urban scenes. During the segmentation stage, a novel region growing approach and a multi-size supervoxel segmentation algorithm robust to noise and varying density are combined to extract effective local shape descriptors. Structural components with physically meaningful labels are generated via structural labelling and clustering. During the classification stage, we consider the structural information at various scales and locations and encode it into a conditional random-field model for unary and pairwise inferences. High-order potentials are also introduced into the conditional random field to eliminate regional label noise. These high-order potentials are defined upon regions independent of connection relationships and can therefore take effect on isolated nodes. Experiments with two MLS datasets of typical urban scenes in Paris and Hong Kong were used to evaluate the performance of the proposed method. Nine and eleven different object classes were recognized from these two datasets with overall accuracies of 97.13% and 95.79%, respectively, indicating the effectiveness of the proposed method of interpreting complex urban scenes from point clouds with large variations in point density. Compared with previous studies on the Paris dataset, our method was able to recognize more classes and obtained a mean F 1 -score of 72.70% of seven common classes, being higher than the best of previous results. [ABSTRACT FROM AUTHOR]