Point Cloud Research Articles

Calibration of the spherical tip radius of Rockwell hardness diamond indenters using a confocal laser scanning microscope

Abstract The precise form of an indenter is an essential part of hardness testing methods. It is therefore incumbent upon a user to ensure that the indenter geometry is calibrated before performing a hardness test. A geometric change of the tip radius by ±5 μm can cause a change of approximately 0.6 units of Rockwell hardness C scale (HRC) in a material with a hardness of 65 HRC. Keeping in mind that the measurement uncertainty is typically of the order of 0.3 HRC, it is critical to know the true value of the tip radius. Typically, tactile methods are used to determine the tip radius of a Rockwell hardness diamond indenter from the measured surface topography. Two main drawbacks of tactile measurements are the long duration of measurement and the limitation in capturing surface features of sizes smaller than the probe tip radius of 2 μm. Both could be overcome by using an optical 3D measurement. Confocal laser scanning microscopes (CLSM) allow a fast contactless 3D-mapping of the surface of Rockwell hardness diamond indenters and can be used to obtain geometric information such as the tip radius. The accuracy of these 3D measurements is still under question. Within this work, some of the influencing factors for fast 3D surface measurement are investigated. Using a CLSM with a 50x objective lens and a numerical aperture of 0.95, typical shape deviations of Rockwell diamond indenters are shown. Furthermore, an improved 3D based point cloud method for the evaluation of the indenter radius is presented. The aim of this paper is to explore the capabilities and limits of CLSM to obtain the 3D surface of a Rockwell hardness diamond indenter in order to calibrate the tip radius and compare their results with measurements from traceable stylus instruments.

Two-stage terrestrial laser scan planning framework for geometric measurement of civil infrastructures

Terrestrial laser scanning plays a crucial role in measuring geometry shapes of civil infrastructures. However, its practical application often encounters challenges such as insufficient model quality and acquisition efficiency. To address these issues, this study proposes a two-stage planning scheme that optimises both scan positioning and single-station scan parameters. Recognising the need to consider both structures and background objects, UAV scanning is employed to capture a three-dimensional point cloud model of the actual scenario. A visibility analysis method based on spherical projection is proposed to enhance computational efficiency of virtual scan simulation. Additionally, an improved particle swarm algorithm is utilised to optimise the scan positioning. During onsite deployment, the scanner’s pose is measured to align structural components with the scanner’s coordinate system, allowing for automated scanning based on occupied angular ranges. The effectiveness of this method is demonstrated through geometry measurements of a small outdoor structure and a suspension bridge.

Automatic geometric digital twin of box girder bridge using a laser-scanned point cloud

Geometric modeling is a pivotal step in creating a digital twin for existing bridge structures. Its deficiency of automation makes geometric modeling step time-consuming and laborious. This paper presents a solution for automatically modeling box girder bridges, including external and internal structures, based on laser-scanned point cloud. The solution includes three vital methods: component segmentation, key points extraction of cross-section, and internal structure reconstruction. The results indicate that the established segmentation model, BCR-Net, exhibited better performance than PointNet++ in component segmentation, as demonstrated by mIoU of 0.9751 on the test set. The mean absolute error on the dimension of the pier, bent cap and external and internal structure of the box girder is 0.27 %, 0.38 %, 0.47 %, and 0.48 %, respectively. It means the proposed methods possessed excellent modeling accuracy while ensuring high efficiency than manual modeling, providing a promising solution for digital twin modeling of bridge structures.

Kinematic–kinetic compliant acetabular cup positioning based on preoperative motion tracking and musculoskeletal modeling for total hip arthroplasty

The invention of the surgical robot enabled accurate component implantation during total hip arthroplasty (THA). However, a preoperative surgical planning methodology is still lacking to determine the acetabular cup alignment considering the patient-specific hip functions during daily activities such as walking. To simultaneously avoid implant edgeloading and impingement, this study established a kinematic–kinetic compliant (KKC) acetabular cup positioning method based on preoperative gait kinematics measurement and musculoskeletal modeling. Computed tomography images around the hip joint and their biomechanical data during gait, including motion tracking and foot–ground reaction forces, were collected. Using the reconstructed pelvic and femur geometries, the patient-specific hip muscle insertions were located in the lower limb musculoskeletal model via point cloud registration. The designed cup orientation has to be within the patient-specific safe zone to prevent implant impingement, and the optimized value selected based on the time-dependent hip joint reaction force to minimize the risk of edgeloading. As a validation of the proposed musculoskeletal model, the predicted lower limb muscle activations for seven patients were correlated with their surface electromyographic measurements, and the computed hip contact force was also in quantitative agreement with data from the literature. However, the designed cup orientations were not always within the well-known Lewinnek safe zone, highlighting the importance of KKC surgical planning based on patient-specific biomechanical evaluations.

Completing 3D point clouds of individual trees using deep learning

Abstract In close‐range remote sensing data collected in a forest, occlusion often causes incomplete or sparse point cloud representations of individual trees, impeding accurate 3D reconstruction of tree architecture and estimation of tree height and volume. Recent developments in deep learning (DL) for 3D data have produced approaches for point cloud completion, which could potentially be applied to trees. We explored the potential of a DL approach to fill gaps in dense point clouds representing the main structures of deciduous trees by applying an existing transformer‐based completion model (PoinTr). Complete point clouds are required as training data, but even dense terrestrial laser scanning (TLS) data sets contain gaps caused by occlusion, making it nearly impossible to acquire such data. We therefore investigated the ability of point cloud completion models trained on a range of synthetic data sets to handle occlusion patterns in real‐world point clouds. Despite the limited data set, we successfully fine‐tuned a general pre‐trained completion model to fill gaps within 1 m3 segments of tree point clouds. Fine‐tuning on synthetic tree data improved the model's ability to complete tree objects compared with training on diverse artificial objects. However, the quality of the predictions was influenced by the level of sophistication of the synthetic data. Our results demonstrate that incorporating even limited real‐world TLS data during training can considerably improve completion results but may introduce additional noise in the predictions. 3D point cloud completion with DL has the potential to improve and fill gaps in point clouds of individual trees, facilitating further steps in the processing and analysis of 3D forest data.

Assessment of Solubility Behavior of a Copolymer in Supercritical CO2 and Organic Solvents: Neural Network Prediction and Statistical Analysis.

In the industrial sector, understanding the behavior of block copolymers in supercritical solvents is crucial. While qualitative agreement with polymer solubility curves has been evaluated using complex theoretical models in many cases, quantitative predictions remain challenging. This study aimed to create a rapid and accurate artificial neural network (ANN) model to predict the lower critical solubility and upper critical solubility space of an atypical block copolymer, poly(styrene-co-octafluoropentyl methacrylate) (PSOM), in different supercritical solvent systems over a wide range of temperatures (51.75-182.05 °C) and high pressure (3.28-200.86 MPa). The experimental data set used in this study included one copolymer, five supercritical solvents, one cosolvent, and one initiator. It consisted of seven unique copolymer-solvent combinations (252 cloud point pressures) used to predict the model quantitatively and qualitatively. To predict the PSOM-solvent interactions, the study considered two different input systems: a six-variable system, a five-variable system, and one target output. Initially, we used a three-layer feed-forward neural network to select the best learning algorithm (Levenberg-Marquardt) from 14 different algorithms, considering one sample PSOM-solvent system. Then, the network topology was optimized by varying hidden neuron numbers from 2 to 80 for all seven PSOM-solvent combination systems. The predicted cloud point pressures were in excellent agreement with the experimental cloud point pressures, confirming the model's accuracy. It is clear from the results of a minimum mean square error (≤1.90 × 10-27) and maximum linear regression R 2 (≥0.99) during training, validation, and testing of all the data sets. Further, the ANN model accuracy was tested by statistical analysis, confirming the model's ability to accurately capture the miscibility regions of polymers, enabling efficient processing of various polymer materials. This data-driven approach facilitates the prediction of coexistence curves for other polymers and complex macromolecular systems.

A New Framework for Generating Indoor 3D Digital Models from Point Clouds

Three-dimensional indoor models have wide applications in fields such as indoor navigation, civil engineering, virtual reality, and so on. With the development of LiDAR technology, automatic reconstruction of indoor models from point clouds has gained significant attention. We propose a new framework for generating indoor 3D digital models from point clouds. The proposed method first generates a room instance map of an indoor scene. Walls are detected and projected onto a horizontal plane to form line segments. These segments are extended, intersected, and, by solving an integer programming problem, line segments are selected to create room polygons. The polygons are converted into a raster image, and image connectivity detection is used to generate a room instance map. Then the roofs of the point cloud are extracted and used to perform an overlap analysis with the generated room instance map to segment the entire roof point cloud, obtaining the roof for each room. Room boundaries are defined by extracting and regularizing the roof point cloud boundaries. Finally, by detecting doors and windows in the scene in two steps, we generate the floor plans and 3D models separately. Experiments with the Giblayout dataset show that our method is robust to clutter and furniture point clouds, achieving high-accuracy models that match real scenes. The mean precision and recall for the floorplans are both 0.93, and the Point–Surface Distance (PSD) and standard deviation of the PSD for the 3D models are 0.044 m and 0.066 m, respectively.

Second-order Spatial Measures Low Overlap Rate Point Cloud Registration Algorithm Based On FPFH Features1

When the sensor dynamically collects point cloud data for object or map reconstruction, the registration effect is poor and reconstruction application is difficult with a too low overlap rate of the collected point cloud data. The reason is that the objects are covered, the sensor rotation angle is too large and the speed of movement is too fast. Because of these problems, this paper proposes a point cloud registration algorithm based on FPFH feature matching, combined with second-order spatial measures. Firstly, using the FPFH feature extraction algorithm, the features of each point are extracted, and then feature matching is performed to generate the set of feature point pairs. Secondly, the second-order spatial measure is used to calculate the set of feature point pairs to obtain the second-order spatial measure matrix scores and sort them. Finally, the dichotomy method is used to find the appropriate second-order spatial measure scores for distinguishing the inner points (points in the overlap region) from the outer points (points that do not belong to the overlap region as well as the mismatched points and some disturbances). The contrast experiments between this algorithm and three common point cloud registration algorithms, FPFH-ICP, 4PCS-ICP, and NDT-ICP, on the Stanford dataset and 3DMatch dataset shows that the registration accuracy of the other algorithms decreases significantly with a low overlap rate. But this algorithm still has a high registration accuracy and is less affected by outliers than the other algorithms. Besides, this algorithm can still maintain a good registration effect on different data sets.

High-accuracy 3D reconstruction of step edge with ray aliasing based on projective parallel single-pixel imaging

Optical 3D measurement technology especially fringe projection profilometry (FPP) is widely utilized in industrial production. However, invalid mixed phases are common when measuring step edges in industrial parts which leads to outliers in reconstructed point cloud and ultimately causes reduction of valid point cloud and dimensional measurement errors. This paper first analyzes reasons for failure of step edge measurement in FPP and illustrates feasibility of parallel single-pixel imaging (PSI) applied to eliminate ray aliasing of step edge. Further for a good balance of accuracy and efficiency, we propose an optimized projective parallel single-pixel imaging (pPSI) method combing with edge detection in wrapped phase generated by pPSI patterns and adaptive bi-direction projection strategy to accurately identify direct correspondence matching points of step edge. Experimental results verified that the proposed method can achieve high accuracy of 3D reconstruction on step edge while taking less measurement time.

Autonomous Vehicles Traversability Mapping Fusing Semantic–Geometric in Off-Road Navigation

This paper proposes an evaluating and mapping methodology of terrain traversability for off-road navigation of autonomous vehicles in unstructured environments. Terrain features are extracted from RGB images and 3D point clouds to create a traversal cost map. The cost map is then employed to plan safe trajectories. Bayesian generalized kernel inference is employed to assess unknown grid attributes due to the sparse raw point cloud data. A Kalman filter also creates density local elevation maps in real time by fusing multiframe information. Consequently, the terrain semantic mapping procedure considers the uncertainty of semantic segmentation and the impact of sensor noise. A Bayesian filter is used to update the surface semantic information in a probabilistic manner. Ultimately, the elevation map is utilized to extract geometric characteristics, which are then integrated with the probabilistic semantic map. This combined map is then used in conjunction with the extended motion primitive planner to plan the most effective trajectory. The experimental results demonstrate that the autonomous vehicles obtain a success rate enhancement ranging from 4.4% to 13.6% and a decrease in trajectory roughness ranging from 5.1% to 35.8% when compared with the most developed outdoor navigation algorithms. Additionally, the autonomous vehicles maintain a terrain surface selection accuracy of over 85% during the navigation process.

Precision estimation of winter wheat crop height and above-ground biomass using unmanned aerial vehicle imagery and oblique photoghraphy point cloud data.

Crop height and above-ground biomass (AGB) serve as crucial indicators for monitoring crop growth and estimating grain yield. Timely and accurate acquisition of wheat crop height and AGB data is paramount for guiding agricultural production. However, traditional data acquisition methods suffer from drawbacks such as time-consuming, laborious and destructive sampling. The current approach to estimating AGB using unmanned aerial vehicles (UAVs) remote sensing relies solely on spectral data, resulting in low accuracy in estimation. This method fails to address the ill-posed inverse problem of mapping from two-dimensional to three-dimensional and issues related to spectral saturation. To overcome these challenges, RGB and multispectral sensors mounted on UAVs were employed to acquire spectral image data. The five-directional oblique photography technique was utilized to construct the three-dimensional point cloud for extracting crop height. This study comparatively analyzed the potential of the mean method and the Accumulated Incremental Height (AIH) method in crop height extraction. Utilizing Vegetation Indices (VIs), AIH and their feature combinations, models including Random Forest Regression (RFR), eXtreme Gradient Boosting (XGBoost), Gradient Boosting Regression Trees (GBRT), Support Vector Regression (SVR) and Ridge Regression (RR) were constructed to estimate winter wheat AGB. The research results indicated that the AIH method performed well in crop height extraction, with minimal differences between 95% AIH and measured crop height values were observed across various growth stages of wheat, yielding R2 ranging from 0.768 to 0.784. Compared to individual features, the combination of multiple features significantly improved the model's estimate accuracy. The incorporation of AIH features helps alleviate the effects of spectral saturation. Coupling VIs with AIH features, the model's R2 increases from 0.694-0.885 with only VIs features to 0.728-0.925. In comparing the performance of five machine learning algorithms, it was discovered that models constructed based on decision trees were superior to other machine learning algorithms. Among them, the RFR algorithm performed optimally, with R2 ranging from 0.9 to 0.93. In conclusion, leveraging multi-source remote sensing data from UAVs with machine learning algorithms overcomes the limitations of traditional crop monitoring methods, offering a technological reference for precision agriculture management and decision-making.

UAV or satellites? How to find the balance between efficiency and accuracy in above ground biomass estimation of artificial young coniferous forest?

The accurate estimation of Above Ground Biomass (AGB) is the basis for plantation forest carbon trading. This study focused on Picea crassifolia artificial plantations, extracting individual tree crown diameters and heights using Unmanned Aerial Vehicles (UAV) data and calculating the individual tree biomass using allometric growth equations. These results were then used to train a satellite image AGB prediction model. In additional, satellite images were resampled to different resolutions to assess the impact of satellite image resolution on model the accuracy. Finally, the model with the highest accuracy among the deep learning algorithms was selected to predicts the AGB within the P. crassifolia plantation forest. The results indicated that the accuracy of single tree crown diameters extracted from P. crassifolia point clouds significantly surpassed those extracted from general point clouds and Crown Height Model (CHM), while the accuracy of the heights extracted from all three sources was similar; RepLKNet outperformed GoogLeNet and ResNet in identifying plantation forest; random forest slightly outperformed XGBoost in the capability of AGB prediction, while the accuracy of the AGB prediction models initially increasd and then decreasd with satellite image resolution, reaching the highest accuracy at a resolution of 50 m. This indicates that the optimal satellite image resolution for estimating the AGB in the study area was affected by scale effects of 50 m. Compared with the combination of satellite data and manual field measurements, the concurrent use of UAVs and satellites offers significant advantages in terms of efficiency and accuracy. UAVs can replace manual sampling for carbon sequestration transactions for plantations.

Distribution and Diversity of Phytoplankton in Lake Alau, Maiduguri, Northeast Nigeria

The production of biodiesel from Thevetia peruviana seed oil using two different catalysts (NaOH and KOH) was evaluated. The Physical and chemical properties after extracting the seed oil were determined and the following results were obtained (density; 0.896g/cm3 , pH; 5.6, oil content; 32.5%, moisture content; 1.5%, specific gravity; 1.33, viscosity; 20mm2 /s, and refractive index; 1.48) and (Acid value; 0.3 mg of KOH/g, free fatty acid; 0.45%, iodine value; 38.5gI/100g, saponification value; was 75mg/KOH) respectively. Different catalysts were prepared by dissolving separately (0.18g. 0.20g and 0.21g) w% of NaOH and another 0.18g, 0.20, 0.21g w% of KOH pellets each in 10ml of methanol to produce concentrations of sodium and potassium methoxide respectively. Furthermore, Yellow Oleander Oil (YOO) was divided into 6 portions (25ml each) and poured into 6 different beakers (100ml) and separately placed on a magnetic stirrer. The amount of biodiesel produced at different concentrations of NaOH catalyst was 20.3ml, 20.1ml, and 22.3ml and the amount of biodiesel produced at different concentration of KOH were 24.2ml, 22.1ml, and 21.0ml. The conversion yield of the biodiesel catalysts was 81.2%, 80.4% and 89% for the NaOH catalyst and 96.8%, 88.4% and 84.0 for the KOH catalyst. The properties of the biodiesel such as flash point were 68℃, pour point was 3℃, cloud point was 7℃, cetane number was 98, the acid value was 0.5mg of KOH/g, density was g/cm3 , Iodine value was 3gI/100g, saponification value was 104, viscosity was 1.5mm2 s, the refractive index was 1.49. The results obtained from the study indicate that the physicochemical properties of the synthesized biodiesel are in agreement with the ASTM standard specifications. Both catalysts gave a considerable yield with KOH having a higher yield than NaOH in almost all the concentrations used. Thevetia peruviana can be considered a highly promising feedstock for biodiesel production.

Safe and Reliable Movement of Fast LiDAR-based Self-driving Vehicle

Classification of objects is an important technique for autonomous ground vehicles to identify a surrounding environment and execute safe path planning. In this paper, a method based on horizontal segmentation is proposed to detect cone-shaped objects in vehicle’s vicinity using a LiDAR sensor. A captured point cloud is divided into five layers based on height information, and the division of detected objects into two groups, cones and others, has been made using classifiers available in MATLAB toolboxes. To separate the classified conical objects into four types used to mark the route, an algorithm for their recognition was developed and used. The proposed solution, verified by navigation experiments in real conditions using an unmanned racing car, has gave good results, i.e., a high rate of cone-shaped objects classification, a short processing time and a low computational load. The performed tests have allowed also to diagnose the causes of incorrect classification of objects. Thus, the experimental results indicated that the approach presented in this work can be used in real time for autonomous, collision-free driving along marked routes.

Prediction of abnormal gait behavior of lower limbs based on depth vision

Abstract As a kind of lower-limb motor assistance device, the intelligent walking aid robot plays an essential role in helping people with lower-limb diseases to carry out rehabilitation walking training. In order to enhance the safety performance of the lower-limb walking aid robot, this study proposes a deep vision-based abnormal lower-limb gait prediction model construction method for the problem of abnormal gait prediction of patients’ lower limbs. The point cloud depth vision technique is used to acquire lower-limb motion data, and a multi-posture angular prediction model is trained using long and short-term memory networks to build a model of the user’s lower-limb posture characteristics during normal walking as well as a real-time lower-limb motion prediction model. The experimental results indicate that the proposed lower-limb abnormal behavior prediction model is able to achieve a 97.4% prediction rate of abnormal lower-limb movements within 150 ms. Additionally, the model demonstrates strong generalization ability in practical applications. This paper proposes further ideas to enhance the safety performance of lower-limb rehabilitation robot use for patients with lower-limb disabilities.

3D Object Detection via Residual SqueezeDet

Three-dimensional object detection is a critical task in computer vision with applications in autonomous driving, robotics, and augmented reality. This paper introduces Residual SqueezeDet, a novel network architecture that enhances the performance of 3D object detection on the KITTI dataset. Building upon the efficient SqueezeDet framework, we propose the Residual Fire module, which incorporates skip connections inspired by ResNet architectures into the original Fire module. This innovation improves gradient flow, enhances feature propagation, and allows for more effective training of deeper networks. Our method leverages point cloud and image-based features, employing a Residual SqueezeDet to effectively capture local and global context. Extensive experiments on the KITTI dataset demonstrate that Residual SqueezeDet significantly outperforms the original SqueezeDet, with particularly notable improvements in challenging scenarios. The proposed model maintains computational efficiency while achieving state-of-the-art performance, making it well-suited for real-time applications in autonomous driving. Our work contributes to the field by providing a more accurate and robust solution for 3D object detection, paving the way for improved perception systems in dynamic environments.

Laser Scanner-Based Hyperboloid Cooling Tower Geometry Inspection: Thickness and Deformation Mapping.

Hyperboloid cooling towers are counted among the largest cast-in-place industrial structures. They are an essential element of cooling systems used in many power plants in service today. Their main structural component, a reinforced-concrete shell in the form of a one-sheet hyperboloid with bidirectional curvature continuity, makes them stand out against other towers and poses very high construction and service requirements. The safe service and adequate durability of the hyperboloid structure are guaranteed by the proper geometric parameters of the reinforced-concrete shell and monitoring of their condition over time. This article presents an original concept for employing terrestrial laser scanning to conduct an end-to-end assessment of the geometric condition of a hyperboloid cooling tower as required by industry standards. The novelty of the proposed solution lies in the use of measurements of the interior of the structure to determine the actual thickness of the hyperboloid shell, which is generally disregarded in geometric measurements of such objects. The proposal involves several strategies and procedures for a reliable verification of the structure's verticality, the detection of signs of ovalisation of the shell, the estimation of the parameters of the structure's theoretical model, and the analysis of the distribution of the thickness and geometric imperfections of the reinforced-concrete shell. The idea behind the method for determining the actual thickness of the shell (including its variation due to repairs and reinforcement operations), which is generally disregarded when measuring the geometry of such structures, is to estimate the distance between point clouds of the internal and external surfaces of the structure using the M3C2 algorithm principle. As a particularly dangerous geometric anomaly of hyperboloid cooling towers, shell ovalisation is detected with an innovative analysis of the bimodality of the frequency distribution of radial deviations in horizontal cross-sections. The concept of a complete assessment of the geometry of a hyperboloid cooling tower was devised and validated using three measurement series of a structure that has been continuously in service for fifty years. The results are consistent with data found in design and service documents. We identified a permanent tilt of the structure's axis to the northeast and geometric imperfections of the hyperboloid shell from -0.125 m to +0.136 m. The results also demonstrated no advancing deformation of the hyperboloid shell over a two-year research period, which is vital for its further use.

Laboratory Tests of Metrological Characteristics of a Non-Repetitive Low-Cost Mobile Handheld Laser Scanner.

The popularity of mobile laser scanning systems as a surveying tool is growing among construction contractors, architects, land surveyors, and urban planners. The user-friendliness and rapid capture of precise and complete data on places and objects make them serious competitors for traditional surveying approaches. Considering the low cost and constantly improving availability of Mobile Laser Scanning (MLS), mainly handheld surveying tools, the measurement possibilities seem unlimited. We conducted a comprehensive investigation into the quality and accuracy of a point cloud generated by a recently marketed low-cost mobile surveying system, the MandEye MLS. The purpose of the study is to conduct exhaustive laboratory tests to determine the actual metrological characteristics of the device. The test facility was the surveying laboratory of the University of Agriculture in Kraków. The results of the MLS measurements (dynamic and static) were juxtaposed with a reference base, a geometric system of reference points in the laboratory, and in relation to a reference point cloud from a higher-class laser scanner: Leica ScanStation P40 TLS. The Authors verified the geometry of the point cloud, technical parameters, and data structure, as well as whether it can be used for surveying and mapping objects by assessing the point cloud density, noise and measurement errors, and detectability of objects in the cloud.

From 3D point‐cloud data to explainable geometric deep learning: State‐of‐the‐art and future challenges

AbstractWe present an exciting journey from 3D point‐cloud data (PCD) to the state of the art in graph neural networks (GNNs) and their evolution with explainable artificial intelligence (XAI), and 3D geometric priors with the human‐in‐the‐loop. We follow a simple definition of a “digital twin,” as a high‐precision, three‐dimensional digital representation of a physical object or environment, captured, for example, by Light Detection and Ranging (LiDAR) technology. After a digression into transforming PCD into images, graphs, combinatorial complexes and hypergraphs, we explore recent developments in geometric deep learning (GDL) and provide insight into the application of these network architectures for analyzing and learning from graph‐structured data. We emphasize the importance of the explainability of these models and recognize that the ability to interpret and validate the results of complex models is a crucial aspect of their wider adoption.This article is categorized under: Technologies > Artificial Intelligence

A Fast Multi-Scale of Distributed Batch-Learning Growing Neural Gas for Multi-Camera 3D Environmental Map Building.

Biologically inspired intelligent methods have been applied to various sensing systems in order to extract features from a huge size of raw sensing data. For example, point cloud data can be applied to human activity recognition, multi-person tracking, and suspicious person detection, but a single RGB-D camera is not enough to perform the above tasks. Therefore, this study propose a 3D environmental map-building method integrating point cloud data measured via multiple RGB-D cameras. First, a fast multi-scale of distributed batch-learning growing neural gas (Fast MS-DBL-GNG) is proposed as a topological feature extraction method in order to reduce computational costs because a single RGB-D camera may output 1 million data. Next, random sample consensus (RANSAC) is applied to integrate two sets of point cloud data using topological features. In order to show the effectiveness of the proposed method, Fast MS-DBL-GNG is applied to perform topological mapping from several point cloud data sets measured in different directions with some overlapping areas included in two images. The experimental results show that the proposed method can extract topological features enough to integrate point cloud data sets, and it runs 14 times faster than the previous GNG method with a 23% reduction in the quantization error. Finally, this paper discuss the advantage and disadvantage of the proposed method through numerical comparison with other methods, and explain future works to improve the proposed method.