Submodular Visual Feature Selection

My final project for the Geometry-Based Vision class is investigating ways to select a subset of visual features from a large feature pool. The motivation is following Luca et al.’s Attention and Anticipation in Fast Visual-Inertial Navigation.

The paper formulates feature selections $S$ from a feature pool $F$ as a submodular problem with the general form: $\max_{S\subset F}\; f(S)\quad \text{subject to} \quad |S| \le \kappa$ where $\kappa$ is the maximum number of features set by the user.

The goal is to achieve accurate state estimation using the minimum resources.

The final project poster can be accessed via the image link: