6DOF Pose estimation for space applications

This thesis presents the methods used to estimate the pose of an object. The pose estimation can be done with two possible approaches, one is using markers and the second is model based. This work is divided in two parts, the first part presents the markers based method and the second part the model based method. The first method is implemented in Python using OpenCV, the second method is implemented in C++ using OpenCV and VISP,it is a modular software package done by INRIA. In order to estimate the position and orientation of the object using the markers based method, it is necessary have at least four points, but in order to increase robustness and achieve good results it is better to use at least six markers. In space applications this method can be limiting, because requires that these markers have to be around the target. Even if, this approach has a great accuracy. In this work this method is implemented with three different kind of markers. The first markers Library has been created freely by choosing some symbols. The second and third has been provided respectively by pyzbar and Aruco Library. The detection of markers in the first application has been done through the contours of markers, in the second and third applications using their library. The pose estimation has been done using OpenCV, in particular given the assumption that all markers are in the same plane, it exploits the 2D points to estimate the pose. The model based method, is more free, because in order to estimate the pose of object does not requires the presence of markers but it is necessary know only the model of the object. In this work is used an circular object with two bolts, the following object has been chosen because it reproduces, on a small scale, a launch adapter ring (LAR) of satellite. It has been used two different cameras, one is the stereo fisheye camera T265 and the second is the depth camera D435, in this way it is possible to see two different methods in order to extract the 3D sets of point clouds. In particular after the first detection of object, the tracking has been done using VISP, and the estimation of pose has been done implemented some algorithms of 3D point cloud registration, like ICP and CPD.