R. Paul Mihail :: Research Opportunities

I am always looking for hard working and motivated students who are interested in computational vision. Below is a list of potential research projects available to my students. Please come see me in my office (Nevins Hall, Room 2119) to discuss your interests and find a good project for you.

Structure from Motion with Plants
Structure from motion (SfM) is the problem of computationally recovering 3D geometry (as a point cloud or mesh model) from a set of 2D images. The pipeline consists of three main steps: feature detection, correspondence computation and bundle adjustment. Image features are used as input to the pipeline, so they can affect the output. Man-made objects have features that are geometrically easier to describe (e.g., corners, straight lines) and easier to match in image pairs under different transformations. Plants have complex geometry, and are often surrounded by other vegetation or otherwise complex background that make problem more difficult. In this project we are investigating novel ways to improve the SfM pipeline with plants.
Plant labeling
Automatic object detection from static imagery is an important problem for machine vision. Current approaches involve automatic detection and computation of discriminative features in local image neighborhoods, followed by a global inference process about the existence and properties of the object of interest. The appearance of the objects influences the choices of features and algorithms to use. The initial focus of this project will be to accurately detect and recognize a few species of plants found in the Lake Louise Research Station. Of particular interest and use by cross-disciplinary researchers is Spanish moss.