Accuracy evaluation of segmentation for high resolution imagery and 3D laser point cloud data

In recent years, point cloud data generated with RGB-D cameras, 3D lasers, and 3D LiDARs have been employed frequently in robotic applications. In indoor environments, RGB-D cameras, which have short-range and can only describe the vicinity of the robots, generally are opted due to their low cost. On the other hand, 3D lasers and LiDARs can capture long-range measurements and generally are used in outdoor applications. In this study, we deal with the segmentation of indoor planar surfaces such as wall, floor, and ceiling via point cloud data. The segmentation methods, which are situated in Point Cloud Library (PCL) were executed with 3D laser point cloud data. The experiments were conducted to evaluate the performance of these methods with the publicly available Fukuoka indoor laser dataset, which has point clouds with different noise levels. The test results were compared in terms of segmentation accuracy and the time elapsed for segmentation. Besides, the general characteristics of each method were discussed. In this way, we revealed the positive and negative aspects of these methods for researchers that plan to apply them to 3D laser point cloud data.

Repairing point cloud holes has become an important problem in the research of 3D laser point cloud data, which ensures the integrity and improves the precision of point cloud data. However, for the point cloud data with non-characteristic holes, the boundary data of point cloud holes cannot be used for repairing. Therefore, this paper introduces photogrammetry technology and analyzes the density of the image point cloud data with the highest precision. The 3D laser point cloud data are first formed into hole data with sharp features. The image data are calculated into six density image point cloud data. Next, the barycenterization Bursa model is used to fine-register the two types of data and to delete the overlapping regions. Then, the cross-section is used to evaluate the precision of the combined point cloud data to get the optimal density. A three-dimensional model is constructed for this data and the original point cloud data, respectively and the surface area method and the deviation method are used to compare them. The experimental results show that the ratio of the areas is less than 0.5%, and the maximum standard deviation is 0.0036 m and the minimum is 0.0015 m.

Since its appearance, airborne laser scanning technology has become more convenient and modern with the continuous progress and development of science and technology. Airborne LiDAR has a very wide range of applications in various fields. It can obtain the three-dimensional coordinates of the measured object comprehensively and with high quality. It has the advantages of high efficiency and high precision in DEM production. In this paper, DEM is produced in an area near Guantao County, Handan City. Firstly, airborne LiDAR is used to collect 3D laser point cloud data, and then software is used to process the point cloud. TerraSolid software is used to perform coordinate conversion and point cloud classification on the point cloud, and finally DEM is produced on the classified ground points. The result is a realistic and intuitive 3D model. The results show that the 3D point cloud data collected by airborne LiDAR can fully and accurately reflect the topographic and geomorphic information, and the software tools and algorithms in the processing process can effectively improve the availability and accuracy of the data, and the generated DEM data can provide important data support for terrain analysis, urban planning, natural resource management and other fields.

The segmentation of crop organs from 3D laser point clouds is an important prerequisite work of crop phenotypic parameters in non-destructive measurement. This paper respectively selected the 3D point cloud data of the rapeseed plant with leaf stage and pod stage as the research materials. A novel normal vector-based method for segmentation of the 3D point cloud is presented. First, a 3D scanner, HandyScan 300, was used to obtain 3D point cloud data. Second, using the voxel-based grid method, the original point cloud data were down-sampled at the premise of keeping the shape of point cloud unchanged. Third, according to the characteristics of the point cloud, the two conditions of the normal vector difference and the Euclidean distance between each point could be merged into two necessary conditions of the current class. Finally, the nearest point was searched with a set of labeled point cloud growth and through each point cloud of European radius until the collection of point cloud and the adjacent candidate was in accordance with the current conditions of the finished classification process. Results showed that the angle difference threshold of the normal vector was [0.91, 0.95]. The segmentation effect of the point cloud data of the leaves of the rapeseed plant was the best, which avoided the problem of misclassification and the appearance of over-segmentation. The angle difference threshold of the normal vector was [0.88, 0.91]. The segmentation effect of the point cloud data of the pod of the rapeseed plant was the best, and the accuracy rate reached 97%. Therefore, the validity and feasibility of the method was verified. Accurate segmentation of the plant organ is another foundation for the nondestructive measurement of the phenotypic parameters in the later stage.

The convex – concave feature of 3D laser point cloud data (3D LPCD) is introduced in fragmentation calculation for blast muck piles in open-pit copper mines to improve the calculation efficiency and accuracy of the blast fragmentation of muck piles (BFMP). First, a supervoxel segmentation method based on point cloud curvature feature (named as CFSS method) is constructed to overcome drawbacks, such as the color difference of the same blast muck pile surface is not evident and has little effect on the 3D LPCD segmentation. Then, a supervoxel clustering method based on boundary extraction (named as BESC method) is constructed to cluster the supervoxel data of the blast muck piles. Finally, on the basis of the supervoxel clustering results, the automatic fragment recognition and fragmentation calculation of the blast muck piles are realized, and the evaluation metrics for the calculation results of the BFMP are introduced. Dexing Copper Mine, the largest open-pit copper mine in China, is considered the study area. Case study results and discussion tests reveal that the ratios of big blocks in the three selected test areas are 0.91%, 0.21%, and 0.83%, and more than 80% of the fragments in the blast muck piles are less than 300 mm, satisfying the actual situation and confirming the effectiveness and accuracy of the proposed method, which is significant for optimizing blast designs and improving mining efficiency in open-pit operations.

Building facade structures form the foundation for 3D model reconstructions, making the extraction of facade structures from 3D point clouds a key research area. A method for extracting the building facade structure from image-based laser point clouds by considering semantic information is proposed. First, point cloud segmentation and clustering are applied to organize the data into distinct planes. Second, semantic images and corresponding semantic image laser point cloud models are generated from each plane. Finally, an enhanced method named as SemColorED extracts the facade structures, and followed by optimization based on building morphology. Evaluation of the method using actual 3D laser point cloud data and the Semantic3D dataset shows improved accuracy, recall, and integrity compared to the current methods.

Purpose This study aims to develop an optimized 3D laser point reconstruction using Descent Gradient algorithm. Precise and accurate reconstruction of 3D laser point cloud of the complex environment/object is a key solution for many industries such as construction, gaming, automobiles, aerial navigation, architecture and automation. A 2D laser scanner along with a servo motor/pan tilt/inertial measurement unit is used for generating 3D point cloud (either environment/object or both) by acquiring the real-time data from sensors. However, while generating the 3D laser point cloud, various problems related to time synchronization problem between laser and servomotor and torque variation in servomotors arise, which causes misalignment in stacking the 2D laser scan for generating the 3D point cloud of the environment. Because of the misalignment in stacking, the 2D laser scan corresponding to the erroneous angular and position information by the servomotor and the 3D laser point cloud become distorted in terms of inconsistency for measuring the dimension of the objects. Design/methodology/approach This paper addresses a modified 3D laser system assembled from a 2D laser scanner coupled with a servomotor (dynamixel motor) for developing an efficient 3D laser point cloud with the implementation of an optimization technique: descent gradient filter (DGT). The proposed approach reduces the cost function (error) in the angular and position coordinates of the servo motor caused because of torque variation and time synchronization, which resulted in enhancing the accuracy in 3D point cloud mapping for the accurate measurement of the object’s dimensions. Findings Various real-world experiments are performed with the proposed DGT filter linked with laser scanner and servomotor and an improvement of 6.5 per cent in measuring the accurate dimension of object is obtained while comparing with conventional approaches for generating a 3D laser point cloud. Originality/value This proposed technique may be applicable for various industrial applications that are based on robotics arms (such as painting, welding and cutting) in the automobile industry, the optimized measurement of object, efficient mobile robot navigation, precise 3D reconstruction of environment/object in construction, architecture applications, airborne applications and aerial navigation.

Quality assessment is one of the essential aspects of the acceptance of transmission line project completion. To resolve the challenge of efficiency and accuracy, we presents a multi-modal data fusion approach established on two new technology, which are the 3D laser point cloud technology and the BIM technology. The scanned 3D laser point cloud data of transmission lines are firstly filtered and classified to build the laser point cloud model; The BIM based 3D model of transmission lines are then converted to the BIM point cloud model; the laser point cloud model and the BIM point cloud model are registered through application of both a coarse registration and a fine registration; Differences between the laser and the BIM point cloud models are assessed for the acceptance of transmission line project. Validation results showed that the multi-modal fusion algorithm had highly satisfactory performance.

Three dimension (3D) point cloud data in fog-filled environments were measured using light detection and ranging (LIDAR). Disaster response robots cannot easily navigate through such environments because this data contain false data and distance errors caused by fog. We propose a method for recognizing and removing fog based on 3D point cloud features and a distance correction method for reducing measurement errors. Laser intensity and geometrical features are used to recognize false data. However, these features are not sufficient to measure a 3D point cloud in fog-filled environments with 6 and 2 m visibility, as misjudgments occur. To reduce misjudgment, laser beam penetration features were added. Support vector machine (SVM) and K-nearest neighbor (KNN) are used to classify point cloud data into ‘fog’ and ‘objects.’ We evaluated our method in heavy fog (6 and 2 m visibility). SVM has a better F-measure than KNN; it is higher than 90% in heavy fog (6 and 2 m visibility). The distance error correction method reduces distance errors in 3D point cloud data by a maximum of 4.6%. A 3D point cloud was successfully measured using LIDAR in a fog-filled environment. Our method’s recall (90.1%) and F-measure (79.4%) confirmed its robustness.

3D digital documentation for buildings has become a necessary tool in preserving them. Heritage buildings are exposed from various kind of threats such as human negligence, natural disaster and weather changes. The fundamental in 3D digital documentation which is the 3D point cloud data has captures great attention and has widely used in many fields due to the availability of laser scanners. The use of laser scanning in engineering surveys is gaining attention due to its advantage of producing high accuracy data. In most situations, it also able to scan the entire required site, thus offers a good potential technique for large-scale applications like for heritage buildings preservation. The data, which consists of high density of points, can be delivered in a short time. However, this causes a massive amount of data generated and hence, it becomes very difficult to be managed. Due to this issue, there are critical needs to have a good method in managing 3D point cloud data to maintain features and visualization of buildings, specially the old and aged ones. This paper will review developed methods in handling these data, concentrating on two specific processes, which are data structure and data filtering. The 3D point cloud data is having a unique representation, thus researchers are no longer concentrating on the usual concepts of data registration, meshing and reconstruction to handle it, but data structure and data filtering are preferred. In data structure, mathematical methods incorporating geometric and topological techniques can be used for studying finite set of points. As most of the data captured contains noises and outliers, filtering is also important and can be treated as one of the processes that can be adapted in handling 3D point cloud data. The implementation of various solutions within these areas are presented in this paper and will be analyzed by emphasizing their contributions. Then, results will be studied to explain the effectiveness of the methods used in handling big point data. Finally, some future work for 3D point cloud handling will be highlighted to conclude this critical review focusing in building data for its preservation.

Many place recognition approaches, which identify previously visited places or locations by matching current sensory data, such as 2D RGB images and 3D point clouds, have been proposed to achieve accurate and robust localization and loop closure detection in global positioning system (GPS)-denied environments. Since visual place recognition (VPR) methods that rely on images captured by camera sensors are highly sensitive to variations in appearance, including changes in lighting, surface color, and shadows, they can lead to poor place recognition accuracy. In contrast, light detection and ranging (LiDAR)-based place recognition (LPR) approaches based on 3D point cloud data that captures the shape and geometric structure of the environment are robust to changes in place appearance and can therefore provide more reliable place recognition results than VPR methods. This work presents an indoor LPR method called PointNetVLAD-based indoor pedestrian localization (PIPL). PIPL is a deep network model that uses PointNetVLAD to learn to extract global descriptors from 3D LiDAR point cloud data. PIPL can recognize places previously visited by a pedestrian using point clouds captured by a low-cost LiDAR sensor on a smartphone in small-scale indoor environments, while PointNetVLAD performs place recognition for vehicles using high-cost LiDAR, GPS, and inertial measurement unit (IMU) sensors in large-scale outdoor areas. For place recognition on 3D point cloud reference maps generated from LiDAR scans, PointNetVLAD exploits the universal transverse mercator (UTM) coordinate system based on GPS and IMU measurements, whereas PIPL uses a virtual coordinate system designed in this study due to the unavailability of GPS indoors. In experiments conducted in campus buildings, PIPL shows significant advantages over NetVLAD (known as a convolutional neural network (CNN)-based VPR method). Particularly in indoor environments with repetitive scenes where geometric structures are preserved and image-based appearance features are sparse or unclear, PIPL achieved 39% higher top-1 accuracy and 10% higher top-3 accuracy compared to NetVLAD. Furthermore, PIPL achieved place recognition accuracy comparable to NetVLAD even with a small number of points in a 3D point cloud and outperformed NetVLAD even with a smaller model training dataset. The experimental results also indicate that PIPL requires over 76% less place retrieval time than NetVLAD while maintaining robust place classification performance.

The Surface Flattening based on Mechanics Revision of the Tunnel 3D Point Cloud Data from Laser Scanner

This paper builds a 3D laser scanner by a one-dimensional pitching rotation pan-tilt and a 2D laser range finder to get the 3D laser point cloud data from the motion environment. Depends on this data, a triangular mesh construction algorithm based on the point cloud matrix is proposed to construct the triangular mesh of the motion environment. Then a triangular plane normal vector clustering algorithm is used to extract the edge feature from the triangular mesh and the mean square deviation is employed as further process to make the edge feature more accurately. The experiment results show that the triangular mesh of motion environment can be constructed effectively and edge feature can be exacted accurately by the algorithms applied above. It lays the foundation of mobile robot autonomous movement in unknown complex environments.

Lithology identification is a critical technology for geological resource exploration and engineering safety assessment. However, traditional methods suffer from insufficient feature representation and low classification accuracy due to challenges such as weathering, vegetation cover, and spectral overlap in complex sedimentary rock regions. This study proposes a hierarchical multi-feature random forest algorithm based on Feature-Preserved Compressive Sampling (FPCS). Using 3D laser point cloud data from the Manas River outcrop in the southern margin of the Junggar Basin as the test area, we integrate graph signal processing and multi-scale feature fusion to construct a high-precision lithology identification model. The FPCS method establishes a geologically adaptive graph model constrained by geodesic distance and gradient-sensitive weighting, employing a three-tier graph filter bank (low-pass, band-pass, and high-pass) to extract macroscopic morphology, interface gradients, and microscopic fracture features of rock layers. A dynamic gated fusion mechanism optimizes multi-level feature weights, significantly improving identification accuracy in lithological transition zones. Experimental results on five million test samples demonstrate an overall accuracy (OA) of 95.6% and a mean accuracy (mAcc) of 94.3%, representing improvements of 36.1% and 20.5%, respectively, over the PointNet model. These findings confirm the robust engineering applicability of the FPCS-based hierarchical multi-feature approach for point cloud lithology identification.

3D Light Detection and Ranging (LiDAR) sensors are closely related to computer vision and deep learning. 3D LiDAR sensors are commonly embedded in smart vehicles to segment humans, cars, trucks, motors, and other objects. However, 3D LiDAR can also be used indoors to predict human poses that are more friendly to a person's privacy because 3D LiDAR does not capture facial images, but it produces data in the form of point clouds. The point cloud produces spatial, geometric, and temporal information which can be used to predict, detect, and classify human poses and activities. The data output from 3D LiDAR, which includes spatial and temporal data, is in PCAP (.pcap) and JSON (.json) formats. The PCAP file contains the sequence frame of the 3D human pose point cloud, and the JSON file contains the metadata. Each human pose class label has one PCAP file and one JSON file. The raw spatio-temporal data must be processed into PCD format as a 3D human pose point cloud dataset for each human pose.The total human pose dataset is 1400 3D point cloud data with PCD format (.pcd) used for the training and testing process in deep learning, consisting of four human pose labels. The label classes are hands-to-the-side, sit-down, squat-down, and stand-up human poses, with each class having 280 3D point cloud data used as training data. While the test data amounted to 280 3D point cloud data. The data collection process uses 3D LiDAR, a tripod, a personal computer/laptop, and a talent, demonstrating basic human poses. The 3D LiDAR used is OS1, a product of Ouster, which has a range of 90–200 m, 128 channels of resolution, and a temperature of -40 – 60° C. For talent, there is one person and male gender in this current shooting. However, in its development, it can also take female or children or elderly talent to enrich the human pose dataset. The talent is between 30 and 40 years old. The distance between the 3D LiDAR and the talent position is 120 cm. Data collection took place from 10:00 a.m. to 1:00 pm. indoors.This dataset is used for human pose prediction using one of the deep learning algorithms, Convolutional Neural Network (CNN). However, the developers can also use other deep learning algorithms such as transformers, Graph Neural Network (GNN), etc.