Real airborne LiDAR data show that the laser intensity on asphalt surface is significantly lower than the one on road marking lines. A Gaussian Surface Matching (GSM) approach is thus developed and presented to determine the parameters of road marking lines and the coordinates of the intersection of two road marking lines as well. Those intersections in the overlapping scanning areas of two neighboring LiDAR strips can be then applied to evaluate the related systematic bias. After systematic bias estimation and correction, the root mean square registration error is reduced from 0.650m to 0.286m in horizontal direction. These intersections are then used as ground control points in aero triangulation (AT). The AT with the precision of ±0.67 pixel is generated.

Mobile lidar system (MLS) which acquires detailed and accurate 3D point clouds along road corridors can assist traditional road inventory work. However, the blind characteristics and the huge amount of point clouds still make it difficult for applications. Therefore, an automatic process for MLS is required. The objective of this study is to develop a novel approach for pole-like road objects extraction from MLS data. The pole-like road objects consist of street light, traffic light, trees, etc. The proposed framework is a model-driven approach. The major work contains three parts: (1) data pre-processing, (2) pole-like object detection, and (3) object classification. In data pre-processing, the raw data are partitioned into several road parts and the building façade aside road are removed to avoid mis-classification. In object detection, a multi-scale segmentation is presented to detect pole-like road objects. Finally, the knowledge-based approach is used to classify pole-like road objects. The experiments demonstrate that the correctness of detection is about 95% and the overall accuracy of classification reaches 72%. The positional accuracy of pole-like object reconstruction is better than 0.1 meters. The results indicate that the method can extract pole-like road objects from MLS data effectively.

Recently, as the continuing improvements in laser sensors with respect to the scanning rate, ranging limit, resolution, accuracy and systemic efficacy, LiDAR has evolved as a major technique for rapid 3-D geo-information acquisition and also led to many new and fascinating applications of a board range of fields. In regard to LiDAR data processing, registration of point clouds scanned from different stations and platforms is one of essential prerequisites for a uniformity of data quality or forming complete scenes. However, the current ways of registration are mostly achieved by using control targets and require a lot of human intervention. Therefore, there still remains a great room for bettering the registration task with less labor consume, high efficiency and good quality. Moreover, how to obtain concrete 3-D information from discrete point clouds is also an important issue. Thus, this paper presents a multiple feature extractor and a novel feature matching approach to realize an automated scheme in feature extraction and pairing among LiDAR point clouds. With the successful demonstrations, the combination of the multiple feature extractor and matcher has been verified as a satisfactory working technique to conduct a variety of LiDAR point clouds

3D scene reconstruction has been applied to broad fields to analyze complex 3D information and display 3D models. The study applies computer vision algorithms, including Structure from Motion (SfM), Clustering Views for Multi-View Stereo (CMVS), Patch-based Multi-View Stereo (PMVS), and Poisson Surface Reconstruction (PSR) to generate 3D scene models with numerous images. This study employs images taken by commercial cameras and smart devices for 3D scene reconstruction through aforesaid algorithms without the need of internal and external parameters. Also, 3D models can be reconstructed in different scales, such as both indoor and outdoor for example. There are four experimental objects containing the helmet and the concrete pillar as small scale models, the pavilion outdoor as a middle scale model, and National Chung Hsing University’s campus as a large scale model to estimate accuracy of focal lengths for 3D models respectively. Finally, reconstructing three different types of concrete pillars and applying constraint rules of a pattern scale was executed to recover the real scale models for volume estimation.

Fire departments already make use of geographic information systems (GIS) to reduce the response time in case of incidents. However, the complex internal structure of buildings on fire may make search and rescue efforts difficult in emergency situations. This study was motivated by the need for a micro GIS to represent and analyze 3D spatial data for fire-fighting simulations of building fires. A method was developed to generate a 3D geometric network model (GNM) from 2D building plans. This model performed network-based analysis within a building. In addition, the applicability of the building information model (BIM) in geospatial environments could provide a 3D visualization environment with sufficient geometric and semantic information about the buildings. Virtual ladder trucks were added in the 3D environment to simulate the positioning and operations of aerial ladders. Finally, the GNM model was implemented to perform fire-fighting simulations based on actual buildings. The proposed method can provide a tool to assist firefighters to find the optimal path within a building, minimize the search time for potential victims, and help firefighters quickly locate their ladder trucks, so as to reduce the response time after an incident.