Search

In the current landscape dominated by Artificial Intelligence, the integration of Machine Learning and Deep Learning within the realm of Cultural Heritage, particularly within architectural contexts, is paramount for the efficient processing and interpretation of point clouds. These advanced methods facilitate automated segmentation and classification, significantly improving both the clarity and practical use of data acquired from laser scanning and photogrammetry. The present study investigates the Palacio de Sástago—a prominent Renaissance palace in Zaragoza, Spain—and introduces a cutting-edge modus operandi for the automated recognition of architectural elements within the palace's inner courtyard. Employing the well-established Random Forest algorithm, implemented in a Python environment, the framework begins with a comprehensive evaluation of the geometric features identified in the LiDAR point cloud. This process employs the Mean Decrease in Impurity metric to evaluate the relevance of each variable. To boost the accuracy and efficiency of the final classifications, the features are refined post-assessment, enhancing both the training phase and the algorithm's later evaluation. The research's findings demonstrate significant potential, supporting advancements in CAD systems and HBIM that will enable more precise, automated modelling of architectural elements, thereby enhancing the accuracy of digital reconstructions and improving conservation planning for heritage sites. [ABSTRACT FROM AUTHOR]

Due to different acquisition time, viewpoint, and sensor noise during the process of point cloud data acquisition, the captured point clouds typically exhibit partial overlapped and contain large amounts of noise and outliers. However, this circumstance tends to diminish the accuracy of point-to-point correspondence searches. Existing point-level methods rely on idealized point-to-point correspondences, which cannot be guaranteed in practical applications. To address above limitations, a noval network based on a geometry transformer and a Gaussian Mixture Model (GMM) is proposed, called GTGMM. Specifically, we formulate the registration problem as the problem of aligning the two Gaussian mixtures, leveraging the advantages of the statistic model and learned robust features to overcome the noise and outliers variants. We utilize a Local Feature Extractor (LFE) to extract structural features of point clouds, while the Transformer encoders establish global relations among the point clouds. Additionally, a geometry transformer network is introduced to capture geometric relations within the point cloud, and overlap scores are learned to reject non-overlapping regions. Utilizing overlap scores, point cloud features, and 3D point cloud coordinates, the matching parameters of GMM to calculate to guide the alignment of two point clouds. Experimental results on synthetic datasets and the real Terracotta Warriors data demonstrate that our method achieves high accuracy and robustness under various registration conditions. [ABSTRACT FROM AUTHOR]

Size reduction of a point cloud or triangulated mesh is an intrinsic part of a three-dimensional (3D) documentation process, reducing the data volume and filtering out erroneous and redundant data obtained during acquisition. Additional reduction has an effect on the geometric accuracy of 3D data compared to the tangible object, and for 3D objects utilized in various cultural heritage applications, the small geometric properties of an object are equally as important as the large ones. In this paper, we investigate several simplification algorithms and various geometric features' relevance to geometric accuracy during the reduction of a 3D object's data size, and whether any of these features have a particular relation to the results of an algorithmic approach. Different simplification algorithms have been applied to several primitive geometric shapes at several reduction stages, and measured values for geometric features and accuracy have been tracked across every stage. We then compute and analyze the correlation between these values to see the effect each algorithm has on different geometries, and whether some of them are better suited for a simplification process based on the geometric features of a 3D object. [ABSTRACT FROM AUTHOR]

Three-dimensional point cloud evaluation is used in photogrammetry to validate and assess the accuracy of data acquisition in order to generate various three-dimensional products. This paper determines the optimal accuracy and correctness of a 3D point cloud produced by a low-cost spherical camera in comparison to the 3D point cloud produced by laser scanner. The fisheye images were captured from a chessboard using a spherical camera, which was calibrated using the commercial Agisoft Metashape software (version 2.1). For this purpose, the results of different calibration methods are compared. In order to achieve data acquisition, multiple images were captured from the inside area of our case study structure (an underpass in Wiesbaden, Germany) in different configurations with the aim of optimal network design for camera location and orientation. The relative orientation was generated from multiple images obtained by removing the point cloud noise. For assessment purposes, the same scene was captured with a laser scanner to generate a metric comparison between the correspondence point cloud and the spherical one. The geometric features of both point clouds were analyzed for a complete geometric quality assessment. In conclusion, this study highlights the promising capabilities of low-cost spherical cameras for capturing and generating high-quality 3D point clouds by conducting a thorough analysis of the geometric features and accuracy assessments of the absolute and relative orientations of the generated clouds. This research demonstrated the applicability of spherical camera-based photogrammetry to challenging structures, such as underpasses with limited space for data acquisition, and achieved a 0.34 RMS re-projection error in the relative orientation step and a ground control point accuracy of nearly 1 mm. Compared to the laser scanner point cloud, the spherical point cloud reached an average distance of 0.05 m and acceptable geometric consistency. [ABSTRACT FROM AUTHOR]

The aim of the paper is to identify an efficient method for transforming the point cloud into parametric objects in the fields of architecture, engineering and construction by four main steps: 3D survey of the structure under investigation, generation of a new point cloud based on feature extraction and identification of suitable threshold values, geometry reconstruction by semi-automatic process performed in Rhinoceros/Grasshopper and BIM implementation. The developed method made it possible to quickly obtain geometries that were very realistic to the original ones as shown in the case study described in the paper. In particular, the application of ShrinkWrap algorithm on the simplify point cloud allowed us to obtain a polygonal mesh model without errors such as holes, non-manifold surfaces, compenetrating surfaces, etc. [ABSTRACT FROM AUTHOR]

The use of a Machine Learning (ML) classification algorithm to classify airborne urban Light Detection And Ranging (LiDAR) point clouds into main classes such as buildings, terrain, and vegetation has been widely accepted. This paper assesses two strategies to enhance the effectiveness of the Deep Learning (DL) classification algorithm. Two ML classification approaches are developed and compared in this context. These approaches utilize the DL Pipeline Network (DLPN), which is tailored to minimize classification errors and maximize accuracy. The geometric features calculated from a point and its neighborhood are analyzed to select the features that will be used in the input layer of the classification algorithm. To evaluate the contribution of the proposed approach, five point-clouds datasets with different urban typologies and ground topography are employed. These point clouds exhibit variations in point density, accuracy, and the type of aircraft used (drone and plane). This diversity in the tested point clouds enables the assessment of the algorithm's efficiency. The obtained high classification accuracy between 89% and 98% confirms the efficacy of the developed algorithm. Finally, the results of the adopted algorithm are compared with both rule-based and ML algorithms, providing insights into the positioning of DL classification algorithms among other strategies suggested in the literature. [ABSTRACT FROM AUTHOR]

In applications based on a three-dimensional point cloud, massive point cloud data often brings processing difficulties. To deal with the problem, many point cloud sampling methods were proposed. But there are still some issues in these methods: (i) lack the consideration of geometric features and (ii) how to train the distribution of projected points by the observed coefficient effectively. This paper introduces a fine-tuned pointnet module, which extracts the geometric features of points and applies the offset-attention mechanism to enhance the feature expression ability. Furthermore, it corrects the positions of simplified points by a mutual supervision loss. The experimental results show our method can improve the effectiveness and robustness of the point cloud sampling. [ABSTRACT FROM AUTHOR]

Monitoring biodiversity in forests is crucial for their management and preservation, especially in light of increasing climatic disturbances. However, traditional methods of surveying forest biodiversity, such as the inventory of tree-related microhabitats (TreMs), are costly and time-consuming. For many years, terrestrial laser scanning (TLS) was the main method for producing highly accurate 3D models of forests. However, with recent advancements in 3D scanning technologies, there are now numerous alternatives available on the market. The aim of this study was to evaluate the performance of four different 3D data acquisition methods, i.e. close-range photogrammetry (CRP), fish-eye photogrammetry (FEP), mobile laser scanning (MLS), and mixed reality depth camera (MRDC), in terms of accuracy and ability to measure biodiversity (TreMs) at tree-stem level, in comparison to TLS. Analysis was performed based on geometric accuracy and point neighbourhood relevance. CRP was the most accurate alternative to TLS for TreM measurement with a median error of 1.5 cm, while FEP provided a good balance between accuracy (median error 1.4 cm) and speed of data collection. Although MLS showed promising results (median error 1.6 cm), noise in the point cloud limited its ability to identify TreMs. MRDC, on the other hand, had lower quality (median error 3.6 cm) and lower point density, making it unsuitable for TreM segmentation. Nevertheless, the study demonstrated the feasibility of augmenting the real world with virtual content at single-tree-stem level using mixed reality technology. Overall, the 3D scanning technologies presented hold great promise for recording the evolution of biodiversity at stem level.

Objectives: Point cloud has no topological structure, current deep learning semantic segmentation algorithm is difficult to capture geometric features implied in irregular points. In addition, the point cloud is in three-dimensional space with a large amount of data size. If we blindly expand the captive field size during extract neighborhood information, it will increase the number of model parameters, which will make model training difficult. Methods: We propose a point cloud semantic segmentation model based on the dilated convolution and combining elementary geometric features such as angle as the model input. First, during feature extraction, basic geometric features such as the relative coordinates, distance and angle between the centroid and the neighboring points are used as the model input to mine the geometric information. Second, in the process of building local neighborhoods, we expand the image dilated convolution operator to point cloud processing, the point cloud dilated operator can expand the receptive field size with no increasing the number of parameters of the model. Finally, the dilated convolution operator, multi-geometric features encoding modules and U-Net architecture are combined to form a complete point cloud semantic segmentation model. Results: The results show that compared with the traditional neighborhood structure, the overall accuracy (OA) of dilated neighborhood structure is increased by 1.4%. Compared with the model that only uses coordinates as input, multi-geometric features encoding module is increased by 10.7%. The final model based on the two proposed algorithms get mean intersection over union and OA are 91.2% and 68.2%, respectively. Conclusions: The dilated neighborhood structure can effectively extract point cloud information in a larger range without increasing the number of model parameters. multigeometric features encoding module can maximize the capture of shape information in the neighborhood. [ABSTRACT FROM AUTHOR]

The semantic segmentation of point clouds has significant applications in fields such as autonomous driving, robot vision, and smart cities. As LiDAR technology continues to develop, point clouds have gradually become the main type of 3D data. However, due to the disordered and scattered nature of point cloud data, it is challenging to effectively segment them semantically. A three-dimensional (3D) shape provides an important means of studying the spatial relationships between different objects and their structures in point clouds. Thus, this paper proposes a semi-supervised semantic segmentation network for point clouds based on 3D shape, which we call SBSNet. This network groups and encodes the geometric information of 3D objects to form shape features. It utilizes an attention mechanism and local information fusion to capture shape context information and calculate the data features. The experimental results showed that the proposed method achieved an overall intersection ratio of 85.3% in the ShapeNet dataset and 90.6% accuracy in the ModelNet40 dataset. Empirically, it showed strong performance on par or even better than state-of-the-art models. [ABSTRACT FROM AUTHOR]

As the technology of high-precision 3D laser scanning becomes increasingly prevalent in the fields of hydraulic building modeling and deformation monitoring, the quality of point clouds plays an increasingly crucial role in data processing. This paper investigates an automatic extraction method of anchor rod points based on geometric features, which focuses on the influence of anchor rod points and mixed pixels in the data of an underground powerhouse of a pumped storage power station on modeling and deformation monitoring during the construction period. This workflow consists of two steps that can automatically extract anchor rod points from high-density point cloud data. Triangular mesh features in the local neighborhood and the parameters of the anchor rods are used to locate the anchor rod in downsampled data, and curvature features are used to extract anchor rod points precisely. The experiment of extracting anchor rods shows that the accuracy of this method of initial identification is 97.2%. Furthermore, precise extraction based on curvature curve fitting is applicable. This method can accurately separate the three types of anchor rods from the dense point cloud on the rough surface of a cavern roof; the false-extraction rate of anchor rod points is about 0.11% to 5.09%. This method can provide high-quality and dependable data sources for the precise registration, modeling and deformation analysis of point clouds in a construction cavern. [ABSTRACT FROM AUTHOR]

Semantic segmentation of mobile LiDAR point clouds is an essential task in many fields such as road network management, mapping, urban planning, and 3D High Definition (HD) city maps for autonomous vehicles. This study presents an approach to improve the evaluation metrics of deep-learning-based point cloud semantic segmentation using 3D geometric features and filter-based feature selection. Information gain (IG), Chi-square (Chi2), and ReliefF algorithms are used to select relevant features. RandLA-Net and Superpoint Grapgh (SPG), the current and effective deep learning networks, were preferred for applying semantic segmentation. RandLA-Net and SPG were fed by adding geometric features in addition to 3D coordinates (x, y, z) directly without any change in the structure of the point clouds. Experiments were carried out on three challenging mobile LiDAR datasets: Toronto3D, SZTAKI-CityMLS, and Paris. As a result of the study, it was demonstrated that the selection of relevant features improved accuracy in all datasets. For RandLA-Net, mean Intersection-over-Union (mIoU) was 70.1% with the features selected with Chi2 in the Toronto3D dataset, 84.1% mIoU was obtained with the features selected with the IG in the SZTAKI-CityMLS dataset, and 55.2% mIoU with the features selected with the IG and ReliefF in the Paris dataset. For SPG, 69.8% mIoU was obtained with Chi2 in the Toronto3D dataset, 77.5% mIoU was obtained with IG in SZTAKI-CityMLS, and 59.0% mIoU was obtained with IG and ReliefF in Paris. [ABSTRACT FROM AUTHOR]

Three-dimensional (3D) branch structures provide vital information for understanding tree phenotypic characteristics and for ecological studies related to carbon sequestration. Light detection and ranging (LiDAR) has been widely applied to capture the 3D structural information of individual trees. Wood–leaf separation and tree skeleton extraction are essential prerequisites for accurately estimating tree attributes (e.g., stem volume, aboveground biomass, and crown characteristics) and representing the tree branch network. Owing to the complex internal branch morphology and intercanopy component occlusion, precise extraction of the tree skeleton from point clouds remains a challenging issue. In this study, we propose an improved approach for extracting tree skeletons on the basis of the geometric features of point clouds. The approach consists of two steps: separation of the wood and leaves, followed by extraction of the tree skeleton. In the first step, the point clouds of the trees are sliced horizontally. The density-based spatial clustering of applications with noise (DBSCAN) algorithm is then employed to cluster each layer of the point clouds and detect the main trunk. Subsequently, random sample consensus (RANSAC) circle feature detection and linear feature constraints are applied to achieve wood–leaf separation. In the second step, the wood point clouds are used to extract the initial tree skeleton via a minimum spanning tree (MST), and the initial tree skeleton is further optimized. Various comparative experiments are conducted on terrestrial-LiDAR-scanned data from nine trees across six species. The results show that the proposed method performs effectively, with overall wood–leaf separation accuracies ranging from 86% to 93%. Additionally, the extracted branch skeleton accurately reflects the natural geometric structure of the trees. The wood points and tree skeletons are further used to estimate tree attributes, demonstrating the potential of our method for the quantitative representation of trees and their ecological characteristics (e.g., carbon sequestration). • Wood and leaf components are effectively separated from point clouds. • An improved approach for extracting tree skeletons from point clouds is proposed. • The extracted branch skeleton accurately represents the natural geometric structure of trees. • Accurate three-dimensional tree attributes can be estimated. [ABSTRACT FROM AUTHOR]

The semantic segmentation of point clouds has significant applications in fields such as autonomous driving, robot vision, and smart cities. As LiDAR technology continues to develop, point clouds have gradually become the main type of 3D data. However, due to the disordered and scattered nature of point cloud data, it is challenging to effectively segment them semantically. A three-dimensional (3D) shape provides an important means of studying the spatial relationships between different objects and their structures in point clouds. Thus, this paper proposes a semi-supervised semantic segmentation network for point clouds based on 3D shape, which we call SBSNet. This network groups and encodes the geometric information of 3D objects to form shape features. It utilizes an attention mechanism and local information fusion to capture shape context information and calculate the data features. The experimental results showed that the proposed method achieved an overall intersection ratio of 85.3% in the ShapeNet dataset and 90.6% accuracy in the ModelNet40 dataset. Empirically, it showed strong performance on par or even better than state-of-the-art models.

This paper proposed a feature descriptor based on quadratic error, which had rotation invariance. It obtained two feature descriptors by extracting the quadratic error of the point and the quadratic error of the neighboring point. It emerged point cloud registration algorithms based on the Gaussian mixture model in an endless stream. The main reason was that probabilistic models had better robustness in terms of noise and outliers. However, this type of method did not perform well for larger-scale rotations. This paper used the quadratic error feature descriptor as a local feature of the Gaussian mixture model to optimize the registration effect in the larger rotation of the Gaussian mixture model, and proposed a dual-feature-based registration strategy to optimize the defects of a single feature. Through experiments, compared with the robust ICP and popular feature-based registration algorithm in terms of registration efficiency and precision, the efficiency of the algorithm is 3 - 4 times that of the robust ICP. In large-scale rotation, the proposed algorithm has good robustness and is superior to the most popular algorithms. [ABSTRACT FROM AUTHOR]

Point clouds are widely used in many fields such as photogrammetry, remote sensing, robotics, documentation, autonomous driving. Information extraction from point clouds is becoming important. The size and complexity of the data require modern methods such as deep learning for information ex-fraction. In the literature, many methods have been developed for the classification of point clouds with deep learning. In this study, the classification problem is defined as sequence classification. Popular recurrent neural network methods (RNN) LSTM and BiLSTM which are popular recurrent neural network (RNN) methods were used to solve this problem. Photogrammetric point clouds produced with UAV images were classified using geometric features and RGB values. Overall classification accuracy decreases when feature spaces consisting of only RGB values or only geometric features are used. The combination of RGB and geometric features increased classification accuracy. Also, with two-way learning, the BiLSTM method has higher accuracy than LSTM. The overall accuracy of 85.68% for BiLSTM and 84.12% for LSTM was obtained [ABSTRACT FROM AUTHOR]

The efficiency and accuracy of point set registration is always a challenge to deal with. In this paper, we propose an improved algorithm using a scale invariant feature extracted. The larger transformation scale and rotation angle cause lower registration accuracy. The initial corresponding point set can be obtained using a scale invariant feature transform (SIFT) operator. In addition, the geometric features of the points are combined to remove the wrong points. The unit quaternion algorithm is used to estimate the best rigid body transformation matrix for precise registration. The experiments show that the registration accuracy increases by 7.99%, while the time consumption decreased by 6.24% in a typical indoor scene. [ABSTRACT FROM AUTHOR]

We present a deep learning framework for efficient large-scale 3D point cloud analysis and classification using the designed feature description matrix (FDM). As the 3D points are unordered in the large-scale scene, and no topology structure can be employed directly for classification and recognition, it is difficult to apply deep neural network directly on 3D point clouds as points cannot be arranged in a fixed order as 2D image pixels. We design a new pipeline for 3D data processing by combining the traditional features extraction method and deep learning method. Our FDM encapsulates the 3D features of the point and can be used as the input of the deep neural network for training and testing. The experiments demonstrate that our method can acquire higher classification accuracy compared with our previous work and other state-of-art works.

3D scene classification has become an important research field in photogrammetry, remote sensing, computer vision and robotics with the widespread usage of 3D point clouds. Point cloud classification, called semantic labeling, semantic segmentation, or semantic classification of point clouds is a challenging topic. Machine learning, on the other hand, is a powerful mathematical tool used to classify 3D point clouds whose content can be significantly complex. In this study, the classification performance of different machine learning algorithms in multiple scales was evaluated. The feature spaces of the points in the point cloud were created using the geometric features generated based on the eigenvalues of the covariance matrix. Eight supervised classification algorithms were tested in four different areas from three datasets (the Dublin City dataset, Vaihingen dataset and Oakland3D dataset). The algorithms were evaluated in terms of overall accuracy, precision, recall, F1 score and process time. The best overall results were obtained for four test areas with different algorithms. Dublin City Area 1 was obtained with Random Forest as 93.12%, Dublin City Area 2 was obtained with a Multilayer Perceptron algorithm as 92.78%, Vaihingen was obtained as 79.71% with Support Vector Machines and Oakland3D with Linear Discriminant Analysis as 97.30%.

Classification and segmentation of point clouds have attracted increasing attention in recent years. On the one hand, it is difficult to extract local features with geometric information. On the other hand, how to select more important features correctly also brings challenges to the research. Therefore, the main challenge in classifying and segmenting the point clouds is how to locate the attentional region. To tackle this challenge, we propose a graph-based neural network with an attention pooling strategy (AGNet). In particular, local feature information can be extracted by constructing a topological structure. Compared to existing methods, AGNet can better extract the spatial information with different distances, and the attentional pooling strategy is capable of selecting the most important features of the topological structure. Therefore, our model can aggregate more information to better represent different point cloud features. We conducted extensive experiments on challenging benchmark datasets including ModelNet40 for object classification, as well as ShapeNet Part and S3DIS for segmentation. Both the quantitative and qualitative experiments demonstrated a consistent advantage for the tasks of point set classification and segmentation. [ABSTRACT FROM AUTHOR]

In this paper, a novel framework for the automatic extraction of road footprints from airborne LiDAR point clouds in urban areas is proposed. The extraction process consisted of three phases: The first phase is to extract road points by using the deep learning model PointNet++, where the features of the input data include not only those selected from raw LiDAR points, such as 3D coordinate values, intensity, etc., but also the digital number (DN) of co-registered images and generated geometric features to describe a strip-like road. Then, the road points from PointNet++ were post-processed based on graph-cut and constrained triangulation irregular networks, where both the commission and omission errors were greatly reduced. Finally, collinearity and width similarity were proposed to estimate the connection probability of road segments, thereby improving the connectivity and completeness of the road network represented by centerlines. Experiments conducted on the Vaihingen data show that the proposed framework outperformed others in terms of completeness and correctness; in addition, some narrower residential streets with 2 m width, which have normally been neglected by previous studies, were extracted. The completeness and the correctness of the extracted road points were 84.7% and 79.7%, respectively, while the completeness and the correctness of the extracted centerlines were 97.0% and 86.3%, respectively. [ABSTRACT FROM AUTHOR]

Three-dimensional reconstruction of cochlear model is commonly used to measure the cochlear contour of patients with high accuracy and configure appropriate hearing aids. To improve the matching accuracy of 3D reconstruction of cochlear model, a triple screening point cloud registration method based on image and geometric features is proposed in this paper. Firstly, extracting the image feature points by SIFT algorithm, then eliminating the mismatched point pairs according to the RANSAC algorithm. Secondly, the average value of the normal feature is introduced to select the feature points with obvious geometric feature region. Thirdly, based on the point cloud fast point feature histogram (FPFH), the matching feature point pairs with image features and geometric features are selected to complete the coarse registration. Finally, iterative closest point (ICP) algorithm is used to achieve fine registration. The proposed algorithm is more comprehensive and effective for finding feature points then ensures the accuracy of registration results. In addition, results of the simulation show that the proposed algorithm reduces both the matching error and the total number of iterations compared with some existing algorithms. [ABSTRACT FROM AUTHOR]

Although visual SLAM (simultaneous localization and mapping) methods obtain very accurate results using optimization of residual errors defined with respect to the matching features, the SLAM systems based on 3-D laser (LiDAR) data commonly employ variants of the iterative closest points algorithm and raw point clouds as the map representation. However, it is possible to extract from point clouds features that are more spatially extended and more meaningful than points: line segments and/or planar patches. In particular, such features provide a natural way to represent human-made environments, such as urban and mixed indoor/outdoor scenes. In this paper, we perform an analysis of the advantages of a LiDAR-based SLAM that employs high-level geometric features in large-scale urban environments. We present a new approach to the LiDAR SLAM that uses planar patches and line segments for map representation and employs factor graph optimization typical to state-of-the-art visual SLAM for the final map and trajectory optimization. The new map structure and matching of features make it possible to implement in our system an efficient loop closure method, which exploits learned descriptors for place recognition and factor graph for optimization. With these improvements, the overall software structure is based on the proven LOAM concept to ensure real-time operation. A series of experiments were performed to compare the proposed solution to the open-source LOAM, considering different approaches to loop closure computation. The results are compared using standard metrics of trajectory accuracy, focusing on the final quality of the estimated trajectory and the consistency of the environment map. With some well-discussed reservations, our results demonstrate the gains due to using the high-level features in the full-optimization approach in the large-scale LiDAR SLAM. [ABSTRACT FROM AUTHOR]

3D scene classification has become an important research field in photogrammetry, remote sensing, computer vision and robotics with the widespread usage of 3D point clouds. Point cloud classification, called semantic labeling, semantic segmentation, or semantic classification of point clouds is a challenging topic. Machine learning, on the other hand, is a powerful mathematical tool used to classify 3D point clouds whose content can be significantly complex. In this study, the classification performance of different machine learning algorithms in multiple scales was evaluated. The feature spaces of the points in the point cloud were created using the geometric features generated based on the eigenvalues of the covariance matrix. Eight supervised classification algorithms were tested in four different areas from three datasets (the Dublin City dataset, Vaihingen dataset and Oakland3D dataset). The algorithms were evaluated in terms of overall accuracy, precision, recall, F1 score and process time. The best overall results were obtained for four test areas with different algorithms. Dublin City Area 1 was obtained with Random Forest as 93.12%, Dublin City Area 2 was obtained with a Multilayer Perceptron algorithm as 92.78%, Vaihingen was obtained as 79.71% with Support Vector Machines and Oakland3D with Linear Discriminant Analysis as 97.30%. [ABSTRACT FROM AUTHOR]

Handheld mobile laser scanning (HMLS) can quickly acquire point cloud data, and has the potential to conduct forest inventory at the plot scale. Considering the problems associated with HMLS data such as large discreteness and difficulty in classification, different classification models were compared in order to realize efficient separation of stem, branch and leaf points from HMLS data. First, the HMLS point cloud was normalized and ground points were removed, then the neighboring points were identified according to three KNN algorithms and eight geometric features were constructed. On this basis, the random forest classifier was used to calculate feature importance and perform dataset training. Finally, the classification accuracy of different KNN algorithms-based models was evaluated. Results showed that the training sample classification accuracy based on the adaptive radius KNN algorithm was the highest (0.9659) among the three KNN algorithms, but its feature calculation time was also longer; The validation accuracy of two test sets was 0.9596 and 0.9201, respectively, which is acceptable, and the misclassification mainly occurred in the branch junction of the canopy. Therefore, the optimal classification model can effectively achieve the classification of stem, branch and leaf points from HMLS point cloud under the premise of comprehensive training. [ABSTRACT FROM AUTHOR]

Mobile Mapping is an efficient technology to acquire spatial data of the environment. The spatial data is fundamental for applications in crisis management, civil engineering or autonomous driving. The extrinsic calibration of the Mobile Mapping System is a decisive factor that affects the quality of the spatial data. Many existing extrinsic calibration approaches require the use of artificial targets in a time-consuming calibration procedure. Moreover, they are usually designed for a specific combination of sensors and are, thus, not universally applicable. We introduce a novel extrinsic self-calibration algorithm, which is fully automatic and completely data-driven. The fundamental assumption of the self-calibration is that the calibration parameters are estimated the best when the derived point cloud represents the real physical circumstances the best. The cost function we use to evaluate this is based on geometric features which rely on the 3D structure tensor derived from the local neighborhood of each point. We compare different cost functions based on geometric features and a cost function based on the Rényi quadratic entropy to evaluate the suitability for the self-calibration. Furthermore, we perform tests of the self-calibration on synthetic and two different real datasets. The real datasets differ in terms of the environment, the scale and the utilized sensors. We show that the self-calibration is able to extrinsically calibrate Mobile Mapping Systems with different combinations of mapping and pose estimation sensors such as a 2D laser scanner to a Motion Capture System and a 3D laser scanner to a stereo camera and ORB-SLAM2. For the first dataset, the parameters estimated by our self-calibration lead to a more accurate point cloud than two comparative approaches. For the second dataset, which has been acquired via a vehicle-based mobile mapping, our self-calibration achieves comparable results to a manually refined reference calibration, while it is universally applicable and fully automated. [ABSTRACT FROM AUTHOR]