SCS Undergraduate Thesis Topics

Student Advisor(s) Thesis Topic
Scott Niekum David Wettergreen Reliable Rock Detection and Classification for Autonomous Science

Current planetary rovers rely heavily on explicit instructions given by operators on Earth, wasting a great deal of bandwidth. An onboard science autonomy system is necessary to give rovers an understanding of science data, so that they may make informed, autonomous decisions, allocate bandwidth intelligently, and respond to new discoveries. A reliable rock detector is the first step in understanding local geology and creating such a system. The detection of objects in natural scenes is an open problem that has been addressed by many computer vision and machine learning researchers. It involves a difficult combination of image processing, the use of subtle visual cues, and domain knowledge about the world. A three stage hybrid architecture is presented that handles data from multiple sources and combines computer vision and machine learning techniques to achieve segmentation, detection, and classification of rocks. Results using ground-truthed data from the Atacama Desert in Chile suggest a successful system design.

Close this window