Projects
Automatic Geomorphic Classification of the Martian Surface -- Many studies of Martian surface starts with manual construction of a geomorphic map that shows spatial distribution of various landforms. This is a slow and labor-intensive process. We study how to automate and thus expedite map production using machine learning. More details in the publications section.
Learning Applied to Ground Robotics (LAGR) -- One of the primary goals of this DARPA program is autonomus navigation of mobile robots in unstructured natural environments. My research involves leveraging computer vision for obstacle detection and scene (trail) classification. Here is a report documenting the use of statistical topic models for measuring similarity between various trails navigated by the mobile robot. The idea is that similar models may be used to classify obstacles in similar trails.
Consensus Segmentation -- Can we combine multiple segmentations of an image to obtain more semantic segmentations? We show that combining multiple k-means clusterings of an image can produce image segmentations which are competitive with state of the art segmentation algorithms. Here is a Tech Report with more details.