Airborne Laser Scanner Based Large Scale Mapping Of The Great Wall Heritage

To protect the ancient Great Wall, government launches the National TechnologySupport Program to establish an IntelliSense system for the presence status of the GreatWall. This paper is part of the project and its task is to build the3-D simulation modelof the Great Wall. After detailed investigation of methods for3-D reconstruction,combined with the specific needs of this project, we select the UAV based airborne lasermeasurement system. However, the accuracy of the system depends on both the positionand attitude information, which is respectively offered by Global Satellite Positioningsystem (GPS) and Inertial Measurement System (INS). As a result, the low accuracy ofcivilian GPS and INS comes to be a critical issue in practical application.To solve this problem, previous studies mainly focused on establishing error modelin specific application, relying on features of structural environment or man-madesetting active labels and so on. Yet combined with requirements of the application, thispaper attempts to explore a common solution which can be applied in non-structural andlarge scale natural environments. In this paper, we set the motion mode ofHovering-Scanning-Flying for the UAV and achieve scans of regions below withmounted3-D laser ranger. For the data obtained, the first step is to correct it locallywith ICP-based (Iterative Closest Points-based) scan matching algorithm. In this paperwe make adaptive improvement for ICP and design an ICP-based scan matchingframework so as to automatically complete all the adjacent cloud points registrationduring the3-D reconstruction. Then a3-D laser measurement data based poseGraph-SLAM algorithm is designed, in whichg2o, an optimization method withsuperior performance, is introduced on basis of the above matching algorithm. Combingit with detecting and eliminating redundant poses automatically, the closed-loopdetection algorithm and some other related strategies, we aim to complete theoptimization and adjustment of the entire3-D model globally.Referred to the actual operating environment, we design two simulationexperiments to verify the effectiveness of the solution. Firstly simulations of the GreatWall are built by detailed shaping. As the target object, the laser rangers scan it to obtainpoint cloud data together with position and attitude information with some errors. Theadjacent laser point clouds are matched with ICP-based scan matching algorithm tocorrect the data locally. Based on this step, pose Graph-SLAM algorithm is applied tooptimize the model globally. Eventually, both steps are validated in the simulationexperiments and desired models are constructed from the above processing.