Structure extraction in urbanized aerial images from a single view using a CNN-based approach

High-level structure (HLS) extraction recovers 3D elements on human-made surfaces (objects, buildings, ground, etc.). There are several approaches to HLS extraction. However, most of these approaches are based on processing two or more images captured from different camera views or on processing 3D data in the form of point clouds extracted from the camera images. In general, 3D point cloud and multiple views approaches have good performance for certain scenes with video sequences or image sequences, but they need sufficient parallax in order to guarantee accuracy. To address this problem, an alternative is to process a single RGB image seeking to interpret areas of the images where the human-made structure may be observed, thus removing parallax dependency, but adding the challenge of having to interpret image ambiguities correctly. Motivated by the latter, we propose a methodology for 3D volumetric structure extraction from a single image. Our strategy is to divide and simplify the 3D structure extraction process. For that, our methodology has three steps. First, the structure recognition step provides the segmentation, location, and delimitation of the urbanized structures in the scene. Second, we propose a graph analysis to classify and locate the boundaries between the different urbanized structures in the scene. Third, we use a proposed CNN and the pinhole camera model to extract the 3D volumetric structure. On the other hand, we evaluate this methodology in synthetic and public datasets.

In recent years, due to the significant advancements in hardware sensors and software technologies, 3D environmental point cloud modeling has gradually been applied in the automation industry, autonomous vehicles, and construction engineering. With the high-precision measurements of 3D LiDAR, its point clouds can clearly reflect the geometric structure and features of the environment, thus enabling the creation of high-density 3D environmental point cloud models. However, due to the enormous quantity of high-density 3D point clouds, storing and processing these 3D data requires a considerable amount of memory and computing time. In light of this, this paper proposes a real-time 3D point cloud environmental contour modeling technique. The study uses the point cloud distribution from the 3D LiDAR body frame point cloud to establish structured edge features, thereby creating a 3D environmental contour point cloud map. Additionally, unstable objects such as vehicles will appear during the mapping process; these specific objects will be regarded as not part of the stable environmental model in this study. To address this issue, the study will further remove these objects from the 3D point cloud through image recognition and LiDAR heterogeneous matching, resulting in a higher quality 3D environmental contour point cloud map. This 3D environmental contour point cloud not only retains the recognizability of the environmental structure but also solves the problems of massive data storage and processing. Moreover, the method proposed in this study can achieve real-time realization without requiring the 3D point cloud to be organized in a structured order, making it applicable to unorganized 3D point cloud LiDAR sensors. Finally, the feasibility of the proposed method in practical applications is also verified through actual experimental data.

Accurate semantic segmentation of 3D point clouds is a long-standing problem in remote sensing and computer vision. Due to the unstructured nature of point clouds, designing deep neural architectures for point cloud semantic segmentation is often not straightforward. In this work, we circumvent this problem by devising a technique to exploit structured neural architectures for unstructured data. In particular, we employ the popular convolutional neural network (CNN) architectures to perform semantic segmentation of LiDAR data. We propose a projection-based scheme that performs an angle-wise slicing of large 3D point clouds and transforms those slices into 2D grids. Accounting for intensity and reflectivity of the LiDAR input, the 2D grid allows us to construct a pseudo image for the point cloud slice. We enhance this image with low-level image processing techniques of normalization, histogram equalization, and decorrelation stretch to suit our ultimate object of semantic segmentation. A large number of images thus generated are used to induce an encoder-decoder CNN model that learns to compute a segmented 2D projection of the scene, which we finally back project to the 3D point cloud. In addition to a novel method, this article also makes a second major contribution of introducing the enhanced version of our large-scale public PC-Urban outdoor dataset which is captured in a civic setup with an Ouster LiDAR sensor. The updated dataset (PC-Urban_V2) provides nearly 8 billion points including over 100 million points labeled for 25 classes of interest. We provide a thorough evaluation of our technique on PC-Urban_V2 and three other public datasets.

The inclusion of thermal infrared (TIR) data in point clouds derived from unmanned aircraft system (UAS) imagery can benefit a variety of applications in which surface temperature and 3D geometry are both important discriminants of feature type and condition. Low resolution and narrow fields of view (FOV) of current consumer-grade TIR cameras on UAS, combined with the lack of sharpness and texture in many image regions, may cause failure or poor results from structure from motion (SfM) photogrammetric software, which has gained widespread use for generating point clouds from UAS imagery. This paper proposes a photogrammetric approach for generating 3D multispectral point clouds utilizing coacquired TIR-RGB images. A 3D point cloud is first generated from the RGB imagery using standard SfM procedures. Then the TIR attributes are assigned to points, where the image coordinates of the points in TIR images are estimated using transformation parameters obtained from co-registration procedures. To obtain RGB-to-TIR transformation parameters, this study tests 3D and 2D co-registration approaches. The latter produces better results due to the challenge of calibrating the TIR camera as required for the 3D approach. This proposed approach is advantageous for generating TIR point clouds without loss of photogrammetric precision compared with solely TIR-based SfM, as the 3D accuracy, point density, and reliability are greatly enhanced.

The study of convolutional neural networks for 3D point clouds is becoming increasingly popular, and the difficulty lies mainly in the disorder and irregularity of point clouds. At present, it is straightforward to propose a convolution operation and perform experimental validation. Although good results are achieved, the principles behind them are not explained—i.e., why this can solve the disorder and irregularity of point clouds—and it is difficult for the researchers to design a point cloud convolution network suitable for their needs. For this phenomenon, we propose a point convolution network framework based on spatial location correspondence. Following the correspondence principle can guide us in designing convolution networks adapted to our needs. We analyzed the intrinsic mathematical nature of the convolution operation, and we argue that the convolution operation remains the same when the spatial location correspondence between the convolution kernel points and the convolution range elements remains unchanged. Guided by this principle, we formulated a general point convolution framework based on spatial location correspondence, which explains how to handle a disordered point cloud. Moreover, we discuss different kinds of correspondence based on spatial location, including M-to-M, M-to-N, and M-to-1 relationships, etc., which explain how to handle the irregularity of point clouds. Finally, we give the example of a point convolution network whose convolution kernel points are generated based on the sample’s covariance matrix distribution according to our framework. Our convolution operation can be applied to various point cloud processing networks. We demonstrated the effectiveness of our framework for point cloud classification and semantic segmentation tasks, achieving competitive results with state-of-the-art networks.

Abstract. This study introduces a novel image to a 3D point-cloud translation method with a conditional generative adversarial network that creates a large-scale 3D point cloud. This can generate supervised point clouds observed via airborne LiDAR from aerial images. The network is composed of an encoder to produce latent features of input images, generator to translate latent features to fake point clouds, and discriminator to classify false or real point clouds. The encoder is a pre-trained ResNet; to overcome the difficulty of generating 3D point clouds in an outdoor scene, we use a FoldingNet with features from ResNet. After a fixed number of iterations, our generator can produce fake point clouds that correspond to the input image. Experimental results show that our network can learn and generate certain point clouds using the data from the 2018 IEEE GRSS Data Fusion Contest.

3-Dimensional reconstruction (3D reconstruction) generates a 3D computer model of a real object or scene from data such as images, it involves many stages and open problems. Existing methods focus on point clouds and reconstructed polygonal mesh within Manhattan-world constrains in urban scenes reconstruction. However, when dealing with structures like steel truss cable-stayed bridges with complex topology (i.e., connectivity and genus), existing methods fail to recover an appealing polygonal mesh from highly unstructured and noisy point clouds. A topology-aware 3D reconstruction method which can obtain high-level structures and low-level shapes is proposed in this paper. A convolutional neural network and point cloud network is designed to encode multi-view images and 3D point cloud into a compact code, which is then decoded into structure layouts (i.e., a hierarchical binary structural parsing tree) and 3D shapes (i.e., leaf nodes on the binary tree) by designing a recursive neural network and a distance field network respectively. These high-level structures and low-level shapes constitute a 3D digital model.

3D vision-based bolt loosening assessment using photogrammetry, deep neural networks, and 3D point-cloud processing

Disaster damage detection through synergistic use of deep learning and 3D point cloud features derived from very high resolution oblique aerial images, and multiple-kernel-learning

ABSTRACTVisual–Linguistic large models (V‐LLMs) have demonstrated strong semantic understanding capabilities in understanding visual open‐world scenes. However, there is limited research on how to effectively integrate these models with traditional deep learning networks for classify and segment 3D point clouds. This paper explores the joint training mechanisms of V‐VLLMs and Convolutional Neural Networks (CNNs), constructing a pluggable module that enhances 3D object classification and component segmentation using 2D images, 3D point clouds, and language descriptions. The idea of this pluggable module is to employ contrastive learning between the generalized features extracted by large models from text and image features extracted by the CNN to enhance the robustness and understanding accuracy of the original CNN‐based point cloud network. For experiments, we validated the effectiveness and accuracy of the proposed pluggable module using multiple point cloud classification frameworks, based on datasets such as ModelNet40, ScanObjectNN, and ShapeNetCore. The experimental results demonstrate that the multimodal contrastive learning pluggable module significantly enhances performance across various deep learning frameworks, resulting in significant enhancements in classification and segmentation precision. Additionally, for the multiscale feature PointNet++ framework, the proposed multiscale contrastive learning mechanism provides further performance gains. The Code is available at https://github.com/winbigqian/MCL‐3D‐VCNN.

Aiming at the problem of 3D point cloud noise affecting the efficiency and precision of human body 3D reconstruction in complex scenes, a 3D point cloud registration denoising method for human motion image using depth learning algorithm is proposed. First, two Kinect sensors are used to collect the three-dimensional data of the human body in the scene, and the spatial alignment under the Bursa linear model is used to pre-process the background point cloud data. The depth image of the point cloud is calculated, and the depth image pair is extracted by the convolutional neural network. Furthermore, the feature difference of the depth image pair is taken as the input of the fully connected network and the point cloud registration parameter is calculated, and the above operation is performed iteratively until the registration error is less than the acceptable threshold. Then, the improved C-means algorithm is used to remove the outlier, the noise is clustered, and the large-scale outlier noise is removed. Finally, the high-frequency information is processed by the depth data bilateral filtering method. The experimental results show that compared with the traditional bilateral filtering algorithm and fuzzy C-means algorithm, the proposed method can effectively remove noise of different scales and maintain good performance on the basis of maintaining human body features. In the point cloud model of A, B, and C, the average error of the proposed method is lower than that of the traditional bilateral filtering algorithm with 15.7%, 15.9%, and 19.8%, respectively, and it is lower than that of the fuzzy C-means algorithm with 25.8%, 26.9%, and 30.2%, respectively.

Converting three-dimensional (3D) point cloud data to two-dimensional (2D) image based on virtual structured-light system is a popular method in 3D data compression because the image format is easier to process by the exiting storing and transmitting methods. When this method is used in the 3D point cloud compression, the quantization error is introduced during the coordinate mapping. To solve this problem, a virtual structured-light 3D point cloud compression algorithm based on geometric reshaping is presented. And a least-square system parameter optimization method is proposed to further improve the data decoding accuracy. In the proposed method, the 3D spatial coordinates are reshaped to a 2D matrix first, and then the 2D matrix is stored as a “Holoimage” by using the parameter optimized virtual structured-light system. The quantization error introduced by coordinate mapping between X and Y coordinates of the 3D point cloud and the 2D image pixel coordinates is suppressed, so the decoding accuracy is improved. In addition, the geometric information of the 3D point cloud is hidden synchronously when the 3D point cloud is compressed, which is of great significance in the copyright protection of the point cloud data. Experiments verify the effectiveness of the proposed algorithm. The decoding root mean square error (RMSE) of the proposed method is decreased by 79.86% on average compared with the traditional one under the same compression ratio, and the compression ratio of the proposed method is 1.96 times bigger than the traditional one under the similar decoding accuracy.

Surface defect identification is a crucial task in many manufacturing systems, including automotive, aircraft, steel rolling, and precast concrete. Although image-based surface defect identification methods have been proposed, these methods usually have two limitations: images may lose partial information, such as depths of surface defects, and their precision is vulnerable to many factors, such as the inspection angle, light, color, noise, etc. Given that a three-dimensional (3D) point cloud can precisely represent the multidimensional structure of surface defects, we aim to detect and classify surface defects using a 3D point cloud. This has two major challenges: (i) the defects are often sparsely distributed over the surface, which makes their features prone to be hidden by the normal surface and (ii) different permutations and transformations of 3D point cloud may represent the same surface, so the proposed model needs to be permutation and transformation invariant. In this paper, a two-step surface defect identification approach is developed to investigate the defects’ patterns in 3D point cloud data. The proposed approach consists of an unsupervised method for defect detection and a multi-view deep learning model for defect classification, which can keep track of the features from both defective and non-defective regions. We prove that the proposed approach is invariant to different permutations and transformations. Two case studies are conducted for defect identification on the surfaces of synthetic aircraft fuselage and the real precast concrete specimen, respectively. The results show that our approach receives the best defect detection and classification accuracy compared with other benchmark methods.

Depth cameras used in visual feedback loops typically produce point clouds or their voxelized images. In the paper, we introduce an original sparse convolutional neural network structure tailored to work with 3D point clouds taken from RGB-D camera. The proposed structure is particularly effective for localization and orientation tasks and operates on the entire scene. We demonstrate its application by using the proposed neural network in a computer vision feedback system to track and control the drone indoors. The visual feedback loop measures the drone’s state and uses the information about its position and orientation to control the drone and allow it to fly along a given trajectory. In the experimental part we show the effectiveness of this approach to control the drone position in indoor spaces. We achieve trajectory-following accuracy comparable to that of the drone’s inbuilt controller’s ability to maintain a fixed position. We compare the proposed solution with a YOLO-based detector and sparse convolutional network to angle prediction operating on cropped and aligned drone point clouds.

Several attempts have been made to grasp three‐dimensional (3D) ground shape from a 3D point cloud generated by aerial vehicles, which help fast situation recognition. However, identifying such objects on the ground from a 3D point cloud, which consists of 3D coordinates and color information, is not straightforward due to the gap between the low‐level point information (coordinates and colors) and high‐level context information (objects). In this paper, we propose a ground object recognition and segmentation method from a geo‐referenced point cloud. Basically, we rely on some existing tools to generate such a point cloud from aerial images, and our method tries to give semantics to each set of clustered points. In our method, firstly, such points that correspond to the ground surface are removed using the elevation data from the Geographical Survey Institute. Next, we apply an interpoint distance‐based clustering and color‐based clustering. Then, such clusters that share some regions are merged to correctly identify a cluster that corresponds to a single object. We have evaluated our method in several experiments in real fields. We have confirmed that our method can remove the ground surface within 20 cm error and can recognize most of the objects.