POINT CLOUD FUSION BETWEEEN AERIAL AND VEHICLE LIDAR
creator
Guangyao, Ma
advisor
Agam, Gady
Because of the increasing requirement of precision in region of 3-D map, we began to use LiDAR to establish a more accurate map. There still exist some problems although we have already made a great progress in this area. One of them, which I tried to process during my thesis study, is that we have two points source - Aerial LiDAR Data( Points gotten by Airplane ) and Vehicle LiDAR Data( Points gotten by Vehicle ) - while both of them have a different density and cannot be merged well. This process - Fusion-is kindly similar to registration, the difference is that the points we would like to merge are generated from different devices and have only few points pairs in the same region. For example, the Aerial LiDAR data has a higher points density in the roofs and ground, but lower in the walls. In the meanwhile, the Vehicle LiDAR data has a lot of points in the walls and ground region. It is beneficial to minimize the difference between these two point sets since the process is necessary for modeling, registration and so on. Therefore, my thesis is to minimize the difference between these two data sources, a procedure of Fusion. The main idea is to read the LiDAR data into data structure of Point Cloud, sample their density to the similar level, and select several corresponding special region pairs( we named these regions -chunks, e.g. Median strip and boundaries of road ) with sufficient interesting points to do fusion. Interesting points indicate the points with one and more special features among all points. And, the algorithm we used to implement the fusion is ICP( Iterative Closet Point Algorithm). Not similar to Registration of Point Cloud, research in the Fusion area is rare. Therefore, the existing algorithms are not well suitable in this project. I deduce some new algorithms during my research since the original ICP Algorithm cannot work well. Both Update Equation and Objective Function are modified. In this thesis, PCL( Point Cloud Library ) is mainly used to implement the basic function, such as nding the nearest points and sampling point cloud, and Eigen library to write the core functions( e.g. Modified Iterative Closest Point Alg ). I also use libLAS library to implement the IO operations and MeshLab to visualize the point cloud after modification.
Submitted by Erma Thomas (thomase@iit.edu) on 2015-09-03T18:11:42Z No. of bitstreams: 1 Thesis_Guangyao_Ma_A20289500.pdf: 2657246 bytes, checksum: ef2af73d363adc21196593f62966a37e (MD5)
Made available in DSpace on 2015-09-03T18:11:42Z (GMT). No. of bitstreams: 1 Thesis_Guangyao_Ma_A20289500.pdf: 2657246 bytes, checksum: ef2af73d363adc21196593f62966a37e (MD5) Previous issue date: 2015-05
M.S. in Computer Science, May 2015
2015
2015-05
http://hdl.handle.net/10560/3530
en
Thesis
born digital
application/pdf
In Copyright
http://rightsstatements.org/page/InC/1.0/
Restricted Access
CS / Computer Science
Illinois Institute of Technology
Affiliated department