Research On The Application Of Improved Sfm In Aerial Photogrammetry

Structure from Motion estimate camera pose (the motion) and reconstruction of scene (the structure) with a set of multiple views images that obtained by motion camera. Since computer vision and photogrammetry techniques are similar,many studies introduced SFM into aerial photogrammetry to solve air triangle and improve the automation of Aerophotogrammetry. However, the field of photogrammetry requires high accuracy, the result of traditional SFM dependent on the accuracy of match. On the other hand, the traditional SFM method uses random sample consensus (RANSAC) algorithm to estimate robustly the fundamental matrix(F) and camera pose. This algorithm use a fixed experience value as a threshold,which cannot guarantee the model parameters are estimated in high accuracy for all data sets, An inappropriate threshold will affect the result of reconstruction. Therefore, the accuracy of results that reconstructed by traditional SFM is difficult to achieve the require of photogrammetry specifications.Aiming above problems, this paper proposes a new SFM process, in order to improve the accuracy of three-dimensional reconstruction. The main work is summarized as follows:1)The method combine the feature based matching (FBM) and area based matching (ABM) with NCC and LSM are proposed. This method uses a hierarchical match, FBM with SIFT start at the top of the image pyramid, then run ABM with improved NCC and LSM layer by layer to adjust the position of match points. By this way, we can improve the accuracy of matching position. In order to improve the estimation precision of the camera parameters, we use these exact match points to estimate the fundamental matrix(F) to improve camera parameter. 2)In the stage of estimating the fundamental matrix and camera pose, this paper proposes a improved RANSAC algorithm which setting a adaptive dynamic threshold to estimate the model parameters accurately. Using this improved algorithm, we can improve the precision of reconstruction.Finally, combining the method presented in this paper, using the image sequence obtain by UAV to complete the estimation of camera parameter and recover the3D information of scene.