Enhancing the 3-D capability of science and engineering cameras on the ExoMars rover

The European Space Agency ExoMars Programme will send a rover to Mars in 2022 equipped with a 2 m drill which will search for evidence of extinct life below the surface. This project is focused on the stereo vision systems of the rover, and how better 3-D information can be extracted from the camera images. This is explored through both experimental and simulation-based approaches. The experimental approach was to build a hardware emulator of the three stereo pairs: the Wide-Angle Cameras (WACs) from the Panoramic Camera instrument (PanCam), and the Navigation Camera (NavCam) and Localisation Camera (LocCam) systems. Point clouds were generated from the emulator data and analysed to determine which combination of cameras yields the best point clouds. Under the limited scope of this study, it was shown that the point cloud made from the four views of the WACs and NavCams performed better than any other combination of these images. The simulation approach was to write a model of image plane quantisation error. By incorporating an occupancy grid mapping approach, this method simulates quantisation error in 3-D for all pixel intersections in 2 or more cameras. Existing methods calculate an approximation of the uncertainty volume, which overestimates this form of error by a factor of 2. For the PanCam WACs, positional uncertainty within the point clouds varies according to location and, by extension, position of the object in WAC images. Depth error increases as a function of depth squared, which agrees with current models. Horizontal error increases linearly as a function of depth. These represent the minimum positional errors in 3-D data products from PanCam because calibration and stereo matching errors will be present. Volumetric scene reconstruction and probabilistic modelling of grid point occupancy represents the two most promising areas for further work for the quantisation error model