DESARROLLO E IMPLEMENTACIÓN DE ALGORITMOS PARA CLASIFICAR DATOS LIDAR EN AREAS URBANAS
[EN] Light detection and ranging (LiDAR) is a remote-sensing technique used to obtain three-dimensional (3D) information of the Earth quickly and accurately. In recent years, LiDAR technology systems have been intensively used in different urban applications such as map updating, communication analy...
| Autor: | |
|---|---|
| Tipo de recurso: | tesis doctoral |
| Fecha de publicación: | 2015 |
| País: | España |
| Institución: | Universitat Politècnica de València (UPV) |
| Repositorio: | RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia |
| Idioma: | español |
| OAI Identifier: | oai:riunet.upv.es:10251/59395 |
| Acceso en línea: | https://riunet.upv.es/handle/10251/59395 |
| Access Level: | acceso abierto |
| Palabra clave: | LiDAR Clasificación Urbano Algoritmos Clasificador angular Segmentación Densificación progresiva Interpolación Terreno Edificios Vegetación INGENIERIA CARTOGRAFICA, GEODESIA Y FOTOGRAMETRIA |
| Sumario: | [EN] Light detection and ranging (LiDAR) is a remote-sensing technique used to obtain three-dimensional (3D) information of the Earth quickly and accurately. In recent years, LiDAR technology systems have been intensively used in different urban applications such as map updating, communication analysis and virtual city modeling. In all applications mentioned is a prerequisite to detect the different objects in the scene. In this regard, one of the most challenging topics is considered to be automatic object classification due to the large variety of natural and man-made objects. The increasing demand for a fast, efficient and automatic algorithm to extract three-dimensional urban features was the motive behind this thesis. Specifically, the objective of the research is the development and implementation of efficient algorithms in the field of classification of objects in complex urban areas with minimal dependence on preset thresholds, in order to achieve a greater degree of automation and greater consistency in the results. The developed algorithms cover all phases of processing and pre-processing of LiDAR data, from the organization of the point cloud in spatial data structures to the neighbourhood definition, filtering and outlier detection and the classification of different entities that make up the scene. Furthermore , the whole pipeline proposed herein is fully automatic and is developed in C++. The proposed classification process detects four entities: bare earth, bridges, buildings and vegetation with small objects. To identify each class different algorithms have been developed and evaluated on two complex urban areas. To filter outliers different algorithms have been developed based on analysis of the neighborhood in terms of distribution of altitudes. In this regard, the best results are achieved by using a multiprocess approach with adaptive thresholds, achieving an overall classification accuracy of 99.9%. Bare earth is classified using a new densification method that combines segmentation techniques and features of other methods such as morphological filters and interpolation-based Methods. The inclusion of additional analysis improves the classification, achieving an average of errors less than 0,5 % for both data test. The classification of bridges is carried out by detecting edges and directional analysis of the structural continuity. Once again, the best results are achieved by using a multiprocess algorithm with adaptive thresholds. To differentiate the buildings from vegetation and other small objects is introduced the new concept of angular classification. The more precise classification is obtained by combining the angular classifier with texture analysis techniques, achieving an average errors in building class in the range of 0,40 % and 1,52 %, depending on the application zone. The experimental results confirm the high accuracy achieved to automatically classify urban objects in complex areas. Furthermore, it is noteworthy that the best results are obtained by multiprocess algorithms with adaptive thresholds or combining algorithms with different approaches. This latter aspect is one of the greatest contributions of the research proposed. |
|---|