Stereo vision–based navigation and attitude determination for CubeSat

Abstract

Recently, vision-based navigation systems for space manipulators have attracted a number of researchers working in this field. The technique could be used to deal with several space missions, including space debris classification and cleaning, small satellite capturing, space docking and formation flying. To generate a more accurate location and attitude, this advanced technique can be combined with other navigation methods, such as inertial measurement unit or global positioning system through a Kalman filter. The stereo vision–based navigation system is presented in this thesis. After the calibration operations to rectify the original image into the standard form, the depth map based on a semi-global matching algorithm is applied to provide a three-dimensional environment in the field of view. Because of the poor performance of this algorithm related to the low resolution and low texture of the object, a feature-based matching algorithm is designed to estimate the model of the CubeSat. This algorithm includes a series of morphological operations and feature extraction methods. A Canny detector can be used to isolate the edges from the images while the edges belong to both the target and the background. Thus, these operations can be applied to extract the edges of the CubeSat from the stereo images, combining the algorithms of the Canny detector and Graph Cut. Further, due to the specific shape of the CubeSat, the crossing points related to the edges can be used as the endpoints to estimate the model for each external plane by reprojection and triangulation. As a result, the attitude and location of the CubeSat can be determined. An experiment has been implemented to verify this design with a stereo camera, a model of CubeSat and the software.