Accurate rough terrain modeling from fused 3D point cloud data

In this paper, we present a new technique for 3D data fusion from two heterogeneous range acquisition devices (i.e., Laser range scanner and Microsoft Kinect) for the extraction of accurate, realistic and rapidly surface reconstruction. First, we present an unsupervised classification algorithm to classify the 3D point cloud data of the terrain into coarser and finer regions. The classification of the 3D point cloud data is done by exploiting the statistical measurement properties of the range dataset. The main merits of the classification method are threshold-freedom and independence from 3D data format and resolution, while preserving characteristic of the terrain details. The 3D point cloud acquired from both the range scanners is transformed into a common reference frame using the principle component algorithm. In the reference frame, the fused point cloud data are obtained by integration of coarser regions data from Kinect and finer regions data from a Laser range scanner. The fused point cloud data eliminate the demerits of the both range sensors by complementing each other. After fusion, we apply Delaunay triangulation algorithm to generate the highly accurate, realistic 3D surface of the terrain. Finally, the experimental results demonstrate the highly robust and precision of the proposed approach.