Abstract

Representation of three dimensional ob jects using a set of ori- ented point pair features has been shown to be effective for ob ject recogni- tion and pose estimation. Combined with an efficient voting scheme on a generalized Hough space, existing approaches achieve good recognition ac- curacy and fast operation. However, the performance of these approaches degrades when the ob jects are (self-)similar or exhibit degeneracies, such as large planar surfaces which are very common in both man made and natural shapes, or due to heavy ob ject and background clutter. We pro- pose a max-margin learning framework to identify discriminative features on the surface of three dimensional ob jects. Our algorithm selects and ranks features according to their importance for the specified task, which leads to improved accuracy and reduced computational cost. In addition, we analyze various grouping and optimization strategies to learn the dis- criminative pair features. We present extensive synthetic and real exper- iments demonstrating the improved results.