Nghiên cứu tối ưu hóa thuật toán tự động phân loại dữ liệu đám mây điểm hỗ trợ xây dựng mô hình 3D thành phố thông minh.Nghiên cứu tối ưu hóa thuật toán tự động phân loại dữ liệu đám mây điểm hỗ trợ xây dựng mô hình 3D thành phố thông minh.Nghiên cứu tối ưu hóa thuật toán tự động phân loại dữ liệu đám mây điểm hỗ trợ xây dựng mô hình 3D thành phố thông minh.Nghiên cứu tối ưu hóa thuật toán tự động phân loại dữ liệu đám mây điểm hỗ trợ xây dựng mô hình 3D thành phố thông minh.Nghiên cứu tối ưu hóa thuật toán tự động phân loại dữ liệu đám mây điểm hỗ trợ xây dựng mô hình 3D thành phố thông minh.Nghiên cứu tối ưu hóa thuật toán tự động phân loại dữ liệu đám mây điểm hỗ trợ xây dựng mô hình 3D thành phố thông minh.Nghiên cứu tối ưu hóa thuật toán tự động phân loại dữ liệu đám mây điểm hỗ trợ xây dựng mô hình 3D thành phố thông minh.Nghiên cứu tối ưu hóa thuật toán tự động phân loại dữ liệu đám mây điểm hỗ trợ xây dựng mô hình 3D thành phố thông minh.1 MINISTRY OF EDUCATION AND TRAINING HANOI UNIVERSITY OF MINING AND GEOLOGY DINH HIEN LE RESEARCH ON OPTIMIZING POINT CLOUD CLASSIFICATION ALGORITHM IN SUPPORTING BUILDING 3D SMART CITY MODEL Major Ge.
1 MINISTRY OF EDUCATION AND TRAINING HANOI UNIVERSITY OF MINING AND GEOLOGY DINH HIEN LE RESEARCH ON OPTIMIZING POINT CLOUD CLASSIFICATION ALGORITHM IN SUPPORTING BUILDING 3D SMART CITY MODEL Major: Geomatics and Mapping Engineering Code: 9520503 SUMMARY OF THE PHD THESIS HANOI – 2023 This thesis was completed at: Department of Photogrammetry and Remote Sensing, Faculty of Geomatics and Land Administration, Hanoi University of Mining and Geology Supervisor Assoc Prof Dr Ngoc Quy Bui, Hanoi University of Mining and Geology Reviewer 1: Assoc Prof Dr Van Anh Tran Hanoi University of Mining and Geology Reviewer 2: Assoc Prof Dr Quang Thanh Bui Vietnam National University, Hanoi Reviewer 3: Assoc Prof Dr Minh Hai Pham Vietnam Institute of Geodesy and Cartography The thesis will be defended at the University-level thesis evaluation committee at Hanoi University of Mining and Geology, at on 2023 The thesis can be found at theses libraries: The Hanoi National Library or The library of Hanoi University of Mining and Geology INTRODUCTION Rationale of the study Surveying technology that collects data from traditional instrumentation measuring single points, now appears instruments that can collect spatial information comprehensively and quickly such as terrestrial laser scanners, mobile laser scanners (mobile mapping), airborne LiDAR scanners on aircraft or on UAVs The common data format of these instruments is 3D point cloud data that carries precise information about geographic coordinates and many other information such as color, intensity, pulse echo, With the advent of 3D point cloud data, the real world is presented in a fully visual manner at the right scale In addition, bigger and bigger volume of 3D point cloud data is now being collected, creating favorable conditions to provide diverse and complete information sources for classifying and building content objects of the model 3D smart city with different LoD (Level of Detail: from LoD0 to LoD4) is attached with attribute information to serve planning applications, urban environment management, space, landscape, Therefore, the point cloud classification will provide an input source for building 3D smart city models However, with a large amount of data, the classification of point cloud data is currently mostly relied on the tools of the commercial software which requires large licensing costs, the parameters and algorithms are encrypted making it impossible for the user to modify to improve the accuracy of the product, At the same time, the manual point cloud classification process consumes a large amount of time and labor costs Therefore, the establishment of an automatic point cloud classification to serve the creation of a smart city 3D model on the basis of research on optimizing algorithms and programs to automatically classify point cloud into different objects Different classes of point cloud data make it possible for us to control data processing, save labor and machine costs, optimize production regarding speed and cost, and The process of building 3D smart city models from point cloud data will make an important contribution to the process of building and developing smart cities Therefore, the PhD student has chosen the topic " Research on optimizing point cloud classification algorithm in supporting building 3D smart city model" to help improve the efficiency of geospatial data production in Vietnam and contribute to the building of the country's future smart city Research aims - Establishing a scientific basis and optimizing the algorithm to automatically classify point cloud data, and from that basis, building a computer program to automatically classify LiDAR point cloud data into different classifications; - Building a workflow to create a smart city 3D model from LiDAR point cloud data Research objects The object of the research is LiDAR point cloud data collected by airborne LiDAR integrated with aerial photography, airborne LiDAR integrated with aerial photography technology and automatic point cloud data classification algorithms Research scope - LiDAR point cloud data collected by airborne LiDAR integrated with aerial photography technology; - Hon Gai area, Ha Long, Quang Ninh Research content Study the theoretical basis, operating principle of LiDAR integration technology and aerial photography and characteristics of LiDAR point cloud data To study the overview of smart city concepts, smart city elements and determine the role of smart city 3D model in building smart city services and applications Researching algorithms to automatically classify point cloud data as a basis for building, optimizing and integrating algorithms to help classify LiDAR point cloud data into different object classes Experimenting with building 3D smart city models in Hon Gai, Ha Long, Quang Ninh areas and evaluating the accuracy of the classification algorithm and the proposed smart city 3D model building process Research methods Synthetic analysis methods; Informatics methods; Methods of investigation, experimental survey; Mapping and GIS methods; Point cloud classification method Scientific and practical significance of the thesis Scientific significance: - The thesis researches and synthesizes the scientific basis of the smart city concept in a systematic way From there, it outlines the role of the smart city 3D model in serving the development of smart city applications, creating a premise for building smart city solutions using geospatial data - The thesis is based on researching and synthesizing the existing automatic classification algorithms of point cloud data, proposing a new set of improved and integrated algorithms, with higher automation and accuracy, helps to classify point cloud data into many different classes and at the same time proposes the workflow of automatically building 3D smart city models from airborne LiDAR data This helps to increase production efficiency, save labor, cost and time Simultaneously, this is also a premise to develop more automated algorithms and applications later Practical significance: - The thesis builds an improved ground point classification algorithm that enhances the accuracy of the automatic search for ground measurement points This helps reduce costs and increase production efficiency in existing mapping and topographic modeling processes - The thesis has optimized the algorithm to automatically classify LiDAR point cloud data, from which, data with large volumes of collected data can later be automatically classified into different object classes, facilitate the development of many other applications - The thesis builds the process of building a 3D smart city model from image data and airborne LiDAR point clouds with high automaticity and accuracy, helping to create a 3D city model dataset in a highly automated way quickly to serve the development of smart city applications, services and solutions, while promoting the application of new technologies in the surveying and mapping industry, contributing to bringing new technology to practical production with the goal of modernization and cost reduction Defense points Defense point 1: The improved automatic ground point classification algorithm and the optimization process combined with the proposed classification thresholds help automatically classify point cloud data into classes with high accuracy and facilitating the construction of LoD2 smart city 3D model Defense point 2: The process of acquiring, processing and classifying LiDAR point cloud data and building 3D smart city models is highly automated, therefore, facilitating the further development of smart city applications New points of the thesis - Researched and optimized a new automatic ground point classification algorithm with higher accuracy than the existing algorithm and a new set of combining algorithms and thresholds that help classifiy point cloud into different classes with high accuracy and automation that help build a detailed LoD2 smart city 3D model - Built a LoD2 3D smart city modeling workflow with high automation 10 Thesis structure The thesis is presented in 135 pages excluding appendices with 10 tables and 59 figures CHAPTER 1: OVERVIEW OF RESEARCH ISSUES 1.1 Smart city overview 1.1.1 In the world According to the definition of D Toppeta in 2010 [12], a smart city is: “A city that combines information technology and Web 2.0 technology with other organizations, design and planning efforts to reduce document volume, speed up administrative processes and help create new solutions and innovative ways to manage city complexes to enhance sustainability and livability” There have been many research institutes and laboratories that have been focusing on research and development of smart city applications such as MIT Senseable Lab [13], Future Cities Laboratory [14], SINTEF Smart Cities [15] ], SMART [16], etc According to United Nations estimates, by 2050, 68% of the world's population will live in cities, so smart city applications will become more popular and practical than ever 1.1.2 In Vietnam Economic development is the main driving force for Vietnam to develop infrastructure, including smart urban infrastructure There is a smart urban planning project along Vo Nguyen Giap street from Nhat Tan bridge to Noi Bai airport in Hanoi This is a super large real estate project that was synchronously planned from the very beginning This project will be the key to smart city development of Hanoi from now to 2030 This is a good signal for the development of Smart City in Vietnam in the near future Along with opportunities there are always challenges The development platform of smart city is technology Currently, Vietnam does not have a complete smart city model Some new cities stop at piloting, testing on a narrow scale Management and operation are also a big challenge when we operate a smart city in reality This experience consists of two main parts: urban management experience and industrial system management experience associated with that city [3] 1.1.3 Components of a smart city 1.1.4 The role of spatial data in smart cities 1.1.5 The role of 3D smart city model Through the overview, it can be seen that the smart city concept encompasses a complex system consisting of many stakeholders, many applications, sensors and IoT devices; Spatial databases play an important role in establishing the interaction between systems, technology and management, as a common platform for data integration and mining Therefore, to develop smart city applications, it is necessary to build a spatial database infrastructure based on a smart city 3D model, allowing the integration of sensors, IoT devices, emulator tools Therefore, the thesis will study and build a 3D city model as a framework for a 3D spatial database to serve as a common foundation for integrating other smart city data with sensors and devices that help building and developing smart city applications, services and solutions 1.1.6 3D smart city model The smart city 3D model concept in the thesis is a geodatabase framework to serve as a common foundation to help integrate data and other sensor sources to serve the development of applications, services, and applications smart city solutions 1.1.7 The Level of detail of 3D smart city model The LoD concept is used to represent the details of a 3D object Based on levels of LoD to reflect the scale of the 3D view, the urban planning simulation problem must follow the standards of urban planning projects, There are levels LoD1, LoD2, LoD3 that will be used in urban spatial simulation as the most suitable, ensuring the functions as well as serving effectively for virtual space analysis problems [4] Due to the characteristics of the airborne LiDAR integrated with aerial image data studied in the thesis and its intended use in a large city-wide scale, the thesis will propose a process to build a LoD2 3D smart city model 1.2 An overview of point cloud data classification methods and algorithms 1.2.1 Overview of point cloud data classification methods 1.2.2 Overview of point cloud data classification algorithms 1.2.3 The situation of research on point cloud classification algorithms in Vietnam LiDAR technology was first deployed in Vietnam in 2006 During that time, domestic units had many research projects for production related to LiDAR point cloud data classification [6][7][8][9] However, these studies only stopped at using commercial software sold with the device to classify LiDAR point clouds The classification results must be checked and orthophoto images used to remove incorrect classification points Recently, there have been more in-depth studies on LiDAR point cloud classification methods such as the PhD thesis of Huu Phuong Nguyen in 2022 [6] improving the EM and MCC algorithms to increase efficiency and automatically classify ground points from LiDAR point cloud data Prior to that, Thanh Huyen Nguyen's master thesis in 2019 [7] studied the method of classifying point clouds using K-Means and MCC algorithms Also in 2022 is Sy Cuong Ngo Sy's doctoral thesis [8] researching using enhanced TIN model development algorithm to increase the efficiency of ground point filtering for LiDAR data 1.3 Overview of the point cloud classification algorithms used in the thesis 1.3.1 Ground Filter Algorithm According to the available studies, the topographic surface filtering algorithms are divided into groups: morphometric-based filtering methods [98, 99, 100, 101], iterative interpolation methods [93, 94] and the increasing density method [104, 105, 106, 107] Among all the topographic surface filtering methods, the method of 12 Hình 2.11 Proposed ground classification algorithm 2.2.2.4 Implementing the improved PTD ground classification algorithm 2.2.2.5 The significance of the improved PTD ground filtering algorithm The improved PTD ground filter algorithm proposed in the thesis has the following advantages compared to Axelsson's PTD algorithm: Allowing to search and distribute ground points more densely in accordance with the development of LiDAR scanning technology when the density of points is increasingly dense and the detail of the data is increasing In addition, the restriction on the starting point is also improved by the filtering of noise points carried out in advance 2.2.3 Classification of asphalt by intensity threshold 2.2.4 Vegetation classification method by elevation threshold and NDVI index 2.2.5 Building classification algorithm In the research of the thesis, the plane-expansion clustering 13 method will be used in combination with other algorithms to classify the house point class The steps of this algorithm are depicted in Figure 2.14 After this process, the layers of roof and wall will be separated, this is an important input material to automatically digitize, build 3D model of the house block and serve the next stages of processing Figure 2.14 Building classification algorithm 2.3 Proposing an automatic optimization process for classifying point cloud data The phD student will integrate many methods and algorithms together, to synthesize into a process to help automatically classify the output point cloud data into different classes as follows: 14 Figure 2.15 The proposed automatic optimization process for classifying point cloud data The whole process above will be built into a computer program that automatically classifies LiDAR point cloud data that was assigned color channels from the output image into different subclasses 2.4 Proposing a workflow to build an automated smart city 3D model from airborne LiDar data 15 Figure 2.16 The workflow to build an automated smart city 3D model from airborne LiDar data CHAPTER 3: BUILDING A SMART CITY 3D MODEL IN HON GAI, HA LONG AREA 16 3.1 Introduction of the study area 3.2 Building a program to classify point cloud data 3.2.1 Purpose and requirements for building a program to classify point cloud data 3.2.2 Overall design of the point cloud data classification program Figure 3.2 Overall design of the point cloud data classification program 3.2.3 Functional design of a point cloud data classification computer program 17 The program HUMG - Point Cloud Classifier includes main sets of functions; Set of file management functions; Set of display functions; Set of digitizing functions ; Set of functions to adjust the viewing angle; Set of point cloud data classification functions; Set of auxiliary tools and Help 3.2.4 Building the point cloud data classification program 3.2.5 Finalizing and testing 3.3 Selection of data processing software 3.4 Experiment of creating 3D city models from airborne LiDAR data The process of building a 3D city model from aerial LiDar imaging and scanning flight data consists of main stages presented in the technological process in Figure 2.16 in Chapter 3.4.1 Data preparation The data used in the thesis is collected by the Leica CityMapper airborne LiDAR & imaging system, including the following preparations: 3.4.1.1 Planning flight mission, base stations and control points 3.4.1.2 Deployment of data capture flight 3.4.2 LiDAR and image data processing Airborne LiDAR and images after the process of scanning and collecting data along with measurement data from the base station will be brought back to the office for data processing steps 3.4.3 Point cloud data classification The 4-color-channel LiDAR point cloud data is classified on the basis of the point cloud classification algorithms proposed in Chapter through the software HUMG - Point Cloud Classifier After that, the automatically classified point cloud will be checked 18 and evaluated using the TerraSolid software suite The point cloud classification process is carried out as shown in Figure 2.15 in chapter The following content will explain each stage in detail 3.4.4 Building 3D city model LiDAR data after it has been classified into separate layers by HUMG software - Point Cloud Classifier, along with the photo data will be included in the TerraSolid software suite to proceed with the steps of building 3D city models (a) (b) (c) Figure 3.21 (a),(b),(c) - 3D model of experimental area city - Hon Gai Ha Long at different locations 3.5 Evaluation of research results 3.5.1 Methods of evaluating results