3D Terrain Reconstruction Based On Unmanned Aerial Vehicle Images

Unmanned aerial vehicle(UAV) had been developed rapidly in recent years. As an important method to obtain space data, getting the high resolution aerial images by UAV is cost saving, flexible and able to detect the high-risk area. However the aerial images can only supply 2D information, leading to the limitation for its applications. As a result, transforming the aerial images to 3D data is inevitable. Topographical map is an important part of aerial images, so realizing the 3D terrain reconstruction is of great significance.In this paper, we put forward a complete 3D terrain reconstruction algorithm based on UAV aerial images. Considering the aerial images contains a few kinds of different terrain and their color difference is significant, we segmented the images with SLIC(simple linear iterative clustering) superpixel algorithm in the first place. In order to improve the efficiency of algorithm, we then fuse the neighboring areas with the improved method of SLIC to reduce superpixel number. Next we used SIFT(Scale Invariant Feature Transform) algorithm to extract the feature points and matched them which will minimize the impact induced by difference in shape and color between images. Then we calculated the depth value of feature points by using the binocular stereo vision theory, and computed their transitions between neighboring images. In the end, we generated the DEM(Digital Terrain Models) data with linear interpolation to reconstruct the 3D terrain.By using two aerial images, three aerial images to reconstruct the 3D terrain, and comparing the result to 3D terrain map in Google Earth, we demonstrated that the proposed algorithm is efficient. The improved SLIC algorithm can realize the region segmentation with less superpixel number which leading to a higher accuracy and lower complexity. The feature points' transition between images established the foundation of large-scale 3D terrain reconstruction based on serial aerial images.