Crown volume is an important tree factor used in forest surveys as a prerequisite for estimating biomass and carbon stocks. This study developed a method for accurately calculating the crown volume of individual trees from vehicle-borne laser scanning (VLS) data using a concave hull by slices method. CloudCompare, an open-source three-dimensional (3D) point cloud and mesh processing software package, was used with VLS data to segment individual trees from which single tree crowns were extracted by identifying the first branch point of the tree. The slice thickness and number to be fitted to the canopy point cloud were adaptively determined based on the change rate in area with height, with the area of each slice calculated using the concave hull algorithm with portions of the crown regarded as truncated cones. The overall volume was then calculated as the sum of all sub-volumes. The proposed method was experimentally validated on 30 urban trees by comparing the crown volumes calculated using the proposed method with those calculated using five existing methods (manual measurement, 3D convex hull, 3D alpha shape, convex hull by slices, and voxel-based). The proposed method produced the smallest average crown volume. Gaps and holes in the point cloud were regarded as part of the crown by the manual measurement, 3D convex hull, and convex hull by slices method, resulting in the calculated volume being higher than the true value; the proposed method reduced this effect. These results indicate that the concave hull by slices method can more effectively calculate the crown volume of a single tree from VLS data.

Road crack segmentation based on high-resolution images is an important task in road service maintenance. The undamaged road surface area is much larger than the damaged area on a highway. This imbalanced situation yields poor road crack segmentation performance for convolutional neural networks. In this paper, we first evaluate the mainstream convolutional neural network structure in the road crack segmentation task. Second, inspired by the second law of thermodynamics, an improved method called a recurrent adaptive network for a pixelwise road crack segmentation task is proposed to solve the extreme imbalance between positive and negative samples. We achieved a flow between precision and recall, similar to the conduction of temperature repetition. During the training process, the recurrent adaptive network (1) dynamically evaluates the degree of imbalance, (2) determines the positive and negative sampling rates, and (3) adjusts the loss weights of positive and negative features. By following these steps, we established a channel between precision and recall and kept them balanced as they flow to each other. A dataset of high-resolution road crack images with annotations (named HRRC) was built from a real road inspection scene. The images in HRRC were collected on a mobile vehicle measurement platform by high-resolution industrial cameras and were carefully labeled at the pixel level. Therefore, this dataset has sufficient data complexity to objectively evaluate the real performance of convolutional neural networks in highway patrol scenes. Our main contribution is a new method of solving the data imbalance problem, and the method of guiding model training by analyzing precision and recall is experimentally demonstrated to be effective. The recurrent adaptive network achieves state-of-the-art performance on this dataset.

Mobile laser scanning can quickly and dynamically obtain a wide range of urban scene point clouds. However, due to factors such as occlusion and field of view limitation, it needs to be supplemented by terrestrial laser scanning. The acquisition methods and data quality of mobile point clouds and terrestrial point clouds are quite different, the target of urban scene point clouds is complex and diverse, and the corresponding feature is difficult to extract, so the point cloud fusion is difficult. To this end, a point cloud registration method of mobile and terrestrial scanning based on the target features of artificial ground objects is proposed. Firstly, the data features of mobile laser scanning point clouds and terrestrial laser scanning point clouds are analyzed, and the point clouds are diluted with equal density. Then, the artificial ground objects are extracted as the registration primitives to reduce the scene complexity, and the features of urban scenes and the features of point cloud eigenvalues and principal curvature attributes are analyzed. Combined with the octree voxel index, the multi-scale key point extraction method is constructed to extract the multi-scale key points of registration primitives. Finally, the key point constraint is used to improve the deficiencies of 4PCS (4-Points Congruent Sets) algorithm and ICP (Iterative Closest Point) algorithm to complete the registration of mobile and terrestrial point clouds in different road scenes. Experiments show that the point cloud registration accuracy can reach 2.6 cm, which provides a feasible method for high precision fusion of multi-platform laser point clouds.

A Method for Extracting Street Trees from Mobile LiDAR Point Clouds