One of the goals of planetary exploration is to cache rock samples for subsequentudreturn to the Earth in future Mars Sample Return missions. Rocks on the Martianudsurface are one of the most interesting science targets for geologists and planetaryudscientists. Hence, it is essential to develop a method for the accurate segmenta-udtion of Martian rocks in Mars images. This thesis introduces a new approach toudsegmenting Mars images captured by the NASA Mars Exploration Rover (MER).udAn improved OTSU and Canny operator are utilized for detecting rock regionsudand their space relations, respectively. The closed contours of detected rocks areudgained by the use of template dilatation edge linking for a given set of images.udThese images have been obtained from MER Navcam and Pancam.udExperimental results of six representative images (with di erent illumination lev-udels, spectral bands and scenes) including a total of 128 rocks are shown. In theseudexperiments qualitative and quantitative comparisons are accomplished. The re-udsults demonstrate that the proposed approach is consistent with human perceptionudand is the best in terms of the average values over the performance indices suchudas Precision, Recall and misclassi cation error in comparison to the existing ap-udproaches. Additionally, a method is proposed for computing the size of a detectedudrock through the stereo triangulation technique. Experimental results also showudthat this proposed method o ers better accuracy than the standard disparity al-udgorithm.udCurrently, science target selection, and whether or not it is possible for a robotudarm to touch the target, is accomplished by human operators and scientists onudthe Earth. The use of onboard autonomy would greatly reduce the human inter-udvention, and it would be advantageous if the rover could evaluate autonomouslyudwhether the robot arm could place an instrument against an identi ed science tar-udget. In this thesis a fuzzy logic-based system is presented to address the problemudof autonomous science target touchability evaluation. The touchability of a poten-udtial science target is assessed in terms of its size (the bounding area of the rock),udSV (the science value of the target), distance (the reachable distance of the armudbetween its base and the science target), and orientation (the angular regions ofudthe arm's shoulder azimuth). In particular, the plane in front of the arm is dividedudinto a number of partitions, which are ranked with the di erent touchability levelsudby the use of a fuzzy rule-based system. Simulations on the rank of science objectudtouchability are carried out, via hardware implementation. Based on the real dataudgathered from the cameras and the Schunk arm experimental results successfullyudverify the validity of the proposed touchability approach and associated softwareudand hardware implementations.
展开▼
