Uav Aerial Image-Based Three-Dimensional Reconstruction Of Outdoor Scenes

Unmanned Aerial Vehicles, especially mini-scale UAVs, are generally and widely used for both civilian and military application due to their safety, high mobility and low-cost. The application of UAVs makes it possible to obtain aerial photos with efficiency.In the meantime, with the development of the information industry, the demands of3D modeling of objects and scenes are growing. Other than modeling by mathematics methods, it is becoming mature to use Non-contact scanners to scan the object and reconstruct the model. And this technology, which has a bright prospect, has been used in archaeology, architecture, precise industrial measurement, object recognition and military.However, traditional3D Modeling methods are with the problems of high costs and inconvenience in transportation and using. Especially for those large scale outdoor scenes it is difficult to model them easily and fast. In recent years, the method which based on computer vision had made great progress. This method can rebuild the object’s geometric profile through images, which is cheap and easy to get. However, traditional image based methods are inefficiency and of poor qualities, due to the difficulty in obtaining high quality images of the outdoor scenes that usually with wide range and large scale.This subject aims at solving these problems, combines UAV gained aerial photos with the method of3D modeling based on serial images together, focus on researching outdoor scene3D reconstruction. Taking the aerial photos’advantage of large scale, wide visual angle and good verisimilitude, this subject tries to establish an efficiency, easy-to-use and low cost3D modeling platform specific for large scale outdoor scene. This platform can be available in many fields such as city digitization or geographical surveying.Here are the issue’s major work and results:1. On the basis of the UAV Aerial Photography’s character, the aerial video sequence has large amount of redundant information, which will delay the matching process and decrease matching accuracy. In order to raise the efficiency and precision of3D reconstruction, extraction of the key frames is applied before3D reconstruction. The filtrate algorithm which is based on several constraints such as baseline width, sharpness and feature point pair constraint can exclude plenty of redundant information and extract valid key frames sequence from the aerial video.2. In the3D reconstruction system, the feature points’ extraction and matching is the precondition of reconstruction and also the most important step in the3D reconstruction process. This issue takes the aerial video’s character of wide baseline and complex changing of view angles into consideration. In considering that the main bodies of the outdoor scenes are usually architectures which have distinct geometrical configuration. Several existing feature matching algorithm are compared and an approved feature detection algorithm based on Line Segment is proved to be sensitive to geometrical features and more efficient.3. There are only few systems that can perform outdoor scene3D reconstruction fast and with integrality and efficiency. This issue will establish a complete reconstruction system with UAV platform and aerial image3D reconstruction. It will apply the "Structure from Motion"3D reconstruction frame to aerial image data, improved and optimized according to the aerial image’s characters.The UAV image acquisition platform and3D reconstruction system is tested and well preformed. This technique is valuable in city digitization, topographic surveying and aerial vehicle automatic control.