Optical Flow Method Based Moving Objects Detection And Attitude Information Estimation Of UAVs

In order to exploit how to improve autonomy of unmanned aircrafts,the thesis conducts related researches on object detection methods in aerial videos and visual based attitude estimation technology by inducing optical flow method,which is one of visual methods.First,the thesis proposes a block matching calculation method of optical flow based on Riemannian metric tensors on the basis of traditional algorithm.Due to the use of the gray-space-time tensor descriptor for images information description and Riemannian metric of tensors and the improved Hausdorff(LTS-HD)distance for the difference measurement between tensors as a replacement of traditional Euclidean distance,it enhances the anti-noise performance of the algorithm.It induces the multi-resolution search strategy for the algorithm available in images with large displacement.The simulation results show the algorithm is more accurate than the traditional algorithms,and is valid for optical flow calculation of images with large displacement and light change.Secondly,the thesis proposes an object detection method based on globe motion estimation and vectors clustering of optical flow.This method begins to extract the local optical flow field after global motion estimation and compensation of image series and then to cluster the local motion vectors for object region of interest positioning.After that,the method segments the real object regions in the region of interest and gets the positions of segmented objects.The simulation results show that the method proposed is valid to detect moving objects from aerial videos captured by unmanned aerial vehicles.Next,a horizon and optical flow based attitude information calculated method is studied in this thesis.This method includes two parts: one is to analysis the relationship between attitude angle and horizon parameters based on camera imaging model and then calculate the roll angle and the pitch angle of aircrafts based on horizon parameters,the other is to utilizes the relationship between the optical flow vector field and the velocity field and extracts the sparse optical flow information near horizon to obtain pitch rate and roll rate estimation of aircrafts.This method is validated with designed semi-physical simulation system.Finally,the thesis builds a simulation system based on the proposed methods on the basis of Simulink Toolbox and X-plane and conducts related simulation tests.The results show that the methods are validated to acquire object information and attitude information for aircrafts to fulfill special tasks.