Recently, LiDAR (Light Detection and Ranging)-based place recognition, has been widely concerned because of its robustness to light conditions, seasonal changes, and viewpoint variations. Unlike most of existing methods which represent the whole point cloud scenes with global descriptors, we treat the LiDAR-based place recognition problem as a scene overlap prediction task and propose an end-to-end overlap prediction network, which consists of a feature learning backbone, a feature enhancement module, and an overlap prediction module. Based on the prediction result for each point, the overlapping ratios between two point clouds are computed and used to predict whether these two point clouds are at the same place. In addition, to promote the computational efficiency and reduce the model complexity, a lightweight feature learning backbone is also adopted. The experiments conducted on the KITTI Odometry dataset demonstrate that the proposed method achieves superior performance compared with state-of-the-art methods. The lightweight method also obtains 2x inference speed with little performance degradation compared with the vanilla method.

As an essential task in remote sensing, disparity estimation of high-resolution stereo images is still confronted with intractable problems due to extremely complex scenes and dynamically changing disparities. Especially in areas containing texture-less regions, repetitive patterns, disparity discontinuities, and occlusions, stereo matching is difficult. Recently, convolutional neural networks have provided a new paradigm for disparity estimation, but it is difficult for current models to consider both accuracy and speed. This paper proposes a novel end-to-end network to overcome the aforementioned obstacles. The proposed network learns stereo matching at dual scales, in which the low one captures coarse-grained information while the high one captures fine-grained information, helpful for matching structures of different scales. Moreiver, we construct cost volumes from negative to positive values to make the network work well for both negative and nonnegative disparities since the disparity varies dramatically in remote sensing stereo images. A 3D encoder-decoder module formed by factorized 3D convolutions is introduced to adaptively learn cost aggregation, which is of high efficiency and able to alleviate the edge-fattening issue at disparity discontinuities and approximate the matching of occlusions. Besides, we use a refinement module that brings in shallow features as guidance to attain high-quality full-resolution disparity maps. The proposed network is compared with several typical models. Experimental results on a challenging dataset demonstrate that our network shows powerful learning and generalization abilities. It achieves convincing performance on both accuracy and efficiency, and improvements of stereo matching in these challenging areas are noteworthy.

Object reconstruction from airborne LiDAR data is a hot topic in photogrammetry and remote sensing. Power fundamental infrastructure monitoring plays a vital role in power transmission safety. This paper proposes a heuristic reconstruction method for power pylons widely used in high voltage transmission systems from airborne LiDAR point cloud, which combines both data-driven and model-driven strategies. Structurally, a power pylon can be decomposed into two parts: the pylon body and head. The reconstruction procedure assembles two parts sequentially: firstly, the pylon body is reconstructed by a data-driven strategy, where a RANSAC-based algorithm is adopted to fit four principal legs; secondly, a model-driven strategy is used to reconstruct the pylon head with the aid of a predefined 3D head model library, where the pylon head’s type is recognized by a shape context algorithm, and their parameters are estimated by a Metropolis–Hastings sampler coupled with a Simulated annealing algorithm. The proposed method has two advantages: (1) optimal strategies are adopted to reconstruct different pylon parts, which are robust to noise and partially missing data; and (2) both the number of parameters and their search space are greatly reduced when estimating the head model’s parameters, as the body reconstruction results information about the original point cloud, and relationships between parameters are used in the pylon head reconstruction process. Experimental results show that the proposed method can efficiently reconstruct power pylons, and the average residual between the reconstructed models and the raw data was smaller than 0.3 m.

Current point cloud registration methods are mainly based on local geometric information and usually ignore the semantic information contained in the scenes. In this article, we treat the point cloud registration problem as a semantic instance matching and registration task and propose a deep semantic graph matching method (DeepSGM) for large-scale outdoor point cloud registration. First, the semantic categorical labels of 3D points are obtained using a semantic segmentation network. The adjacent points with the same category labels are then clustered together using the Euclidean clustering algorithm to obtain the semantic instances, which are represented by three kinds of attributes including spatial location information, semantic categorical information, and global geometric shape information. Second, the semantic adjacency graph is constructed based on the spatial adjacency relations of semantic instances. To fully explore the topological structures between semantic instances in the same scene and across different scenes, the spatial distribution features and the semantic categorical features are learned with graph convolutional networks (GCNs), and the global geometric shape features are learned with a PointNet-like network. These three kinds of features are further enhanced with the self-attention and cross-attention mechanisms. Third, the semantic instance matching is formulated as an optimal transport problem, and solved through an optimal matching layer. Finally, the geometric transformation matrix between two point clouds is first estimated by the singular value decomposition (SVD) algorithm and then refined by the ICP algorithm. Experimental results conducted on the KITTI Odometry dataset demonstrate that the proposed method improves the registration performance and outperforms various state-of-the-art methods.

The security of high-voltage power transmission corridors is significantly vital to the national economy and daily life. With its rapid development, LiDAR (Light Detection and Ranging) technology has been widely applied in the inspection of transmission lines. As the basis of potential hazard detection, a robust and precise power line model is a necessary requirement for rapid and correct clearance. Thus, this paper proposes a novel method for high-voltage bundle conductor reconstruction, which can precisely reconstruct each sub-conductor. First, points in high-voltage power transmission corridors are detected and classified into four categories; second, for classified power lines, single power line spans are extracted, and bundle conductors are identified by analyzing the single spans’ fitting residuals; and then, each sub-conductor of bundle conductors is extracted by a projected dichotomy method on the XOY and XOZ planes, respectively; finally, a double-RANSAC (random sample consensus)-based algorithm was introduced to reconstruct each power line. The proposed method makes use of the distribution of bundle conductors in high-voltage transmission lines, and our experiments showed that it could preferably reconstruct the real structure of bundle conductors robustly with a high precision better than 0.2 m.

3D surface matching by local feature descriptors is a fundamental task in 3D object registration, recognition and retrieval. In view of 2D projected descriptors’ information compression and 3D directly voxelized descriptors’ sensitivity to density variation and boundary, this paper firstly proposes a voxel-based buffer-weighted binary descriptor, named VBBD. After local surfaces of detected keypoints are voxelized, each voxel’s buffer region is established, and the buffer-weighted Gaussian kernel density is calculated. If the current voxel’s buffer-weighted density is larger than its local surface’s average density, the voxel is labelled 1, otherwise it is 0. The proposed descriptor has several merits (1) direct acquisition of 3D information without projection, the neighbor information is less compressed. (2) voxels are labelled and binarized according to the buffer-weighted density, which improves the robustness to boundary effect, noise and density variation. Based on the VBBD, a global optimal surface matching method based on a KuhnMunkres (KM) algorithm is adopted. The strategy has several advantages (1) by calculating the descriptors of large-scaled surfaces which are down sampled, the number of point cloud is reduced, and meanwhile, large-scaled information is obtained; (2) the KM algorithm is adopted to find matching pairs to achieve final maximum weight sum of all matching pairs, which can efficiently avoid local optimum. Experimental results show that, compared with other state-of-the-art descriptors, VBBD has better descriptiveness and robustness, and the surface matching strategy by VBBD can achieve both high recall and precision.