Uncertain Reasoning and Learning for 3-D Object Detection and Description
Abstract
Acquiring 3-D object description from a single or multiple images
has been a key goal of computer vision. Building detection and
description is a practical application for this problem. In this
proposal, I follow a hypothesize and verify paradigm, where decision
making at various stages is important. Most of the previous work has
been focused on feature grouping, and the decisions were usually
made by ad hoc operators. In this proposal, I show experiments of
applying uncertain reasoning and learning for hypothesis
verification. First experiment with monocular building detection and
description system verifies the idea that the uncertain reasoning
and learning will bring better result with smaller efforts for
tuning the parameters. Based on these results I apply Bayesian
inference to the multi-view system, where the number of evidence
inputs varies according to the number of views. I propose a dynamic
Bayesian network to deal with this situation. I also introduce a
novel way of designing a Bayesian network structure. In the
experimental results, the proposed method shows a superior
performance to that of a state-of-art classifier. Finally, I propose
a future a research for the multi-layered polygonal rooftop building
detection and description. I show some preliminary results and
discuss remaining scientific challenges.
On-line references