Survey the Performance of UAV Images to Generation a Digital Surface Model

Recently, due to the predominance in terms of the time and the cost, the drones are dramatically developing and their applications getting more widespread in various fields such as earth science, natural resources, environmental studies and urban planning. However, applying such drones suffers from several disadvantages, inter alia, the imaging unstable geometry, not using metric cameras as well as images’ variable parameters. In this study, computer vision science techniques for processing prepared images of the drone were precisely investigated and the performance of these images and techniques for producing a surface Elevation model with a high spatial resolution was investigated. Due to mentioned problems for the imaging and the absence of internal and external justifying parameters, the SIFT operator was used in order to extract nodal points in the shared areas of images. The relative and absolute justification of images are performed by using nodal points and land control points, respectively. The cloud of dense points was created through a dense three-dimensional matching algorithm and by using triangulated irregular interpolation methods area digital surfaces are formed. Finally, by using land checkpoints, the accuracy of the georeferencing and the final surface digital model were assessed. 5853 nodal points were automatically extracted by the SIFT operator and images were georeferenced with a surface accuracy of 21/12 cm. Then, by creating millions of points and triangulated irregular network, the height digital model with a height accuracy of 19/25 cm was created. Results of the current study indicate that due to the imaging unstable geometry and variable parameters for images, the combination of the SIFT algorithm with computer vision techniques in processing the UAV images in order to extract nodal points and to precisely justify images, are necessary.