Roof detection on 3D Mesh Building Model Generate from High-Resolution Oblique Image

Abstract

Nowadays, the trend of using small aerial equipment such as, Unmanned Aerial Vehicle (UAV) mounting the oblique cameras is high efficiency to collect the 3D data, because it's can fly close to the object and has better accessibility to narrow space to capture the detailed information to ensure that all of the building in the area were captured from every single side of the whole building. Especially for the rooftops of the building, which is the main object of our study. The advantage of UAV is cheap for the operation in each time, more flexible, and short production cycle, when compared with the traditional acquisition method, and especially, to generate the high-resolution 3D models. Three-dimensional virtual building models with the high-level detailed information is an important description of the city areas, are wildly used in various fields, such as city plan, design, and solar energy area, helping in their decision process, and building information model (BIM), etc.

However, many applications require a building model with a topologic surface structure rather than mesh data, for the mesh data cannot provide semantic information about the roof and façade. These rooftops surface information can be extracted by using the digital geometry processing algorithm and most of the automatic segmentation methods, especially for the architecturally sophisticated building.

In order to segment rooftops and surface automatically from 3D mesh data generated from oblique images, This dissertation introduces an efficient segmentation method base on the Region Growing algorithm, which is an innovative, robust, and efficient method for three-dimensional building rooftops segmentation. Demonstrated a segmentation method aimed at breaking down the rooftops from the building structure by using the efficiency of triangle mesh derived from UAV. The materials as aerial images were generated to 3D mesh models by using Get3D software. Basically, curvature estimation calculated for all vertices on the surface mesh and classified into several clusters. Then, the use of discrete curvature is calculated for each vertex on the triangle surface. Then vertices are classified into the clusters according to the principal curvature based on values of K-mean clustering. Afterward, the Region Growing algorithm is aimed at breaking down the structure in the meaningful of sub-component. We proposed it can detect opening boundary even on rooftop of the Building, by the mechanism of region growth used to extract the connected regions, starting with a seed region on surface data and grow up to the neighbor when neighbors seed are satisfied in some condition. The growing process is continuing until the vertices are met. Lastly, the used region merging approach is aimed to reduce the over-segmentation result from the growing clusters.

The result illustrated a structure of the roof plane with topology in each step of detection, the efficiency and precision of the proposed algorithm in terms of computation time, extraction accuracy. The rooftops were segmenting from Region Growing algorithm base curvature estimation, it can display the shape and structure that segmented from the mesh models. The integration of both techniques is significantly and advantageous from the detection and segmentation of triangle mesh data. However, the rooftops with topologic surface structure can provide the semantic information can be applied to various fields such as solar roof energy or city plan.

-