Mirrors often cause ambiguities in scene understanding. For example, when a robot sees a corridor reflected from a large mirror, it may not be able to tell it's a reflected scene and wrongly navigate into it. This ambiguity is particularly a problem when using RGB (image) information only. On the other hand, depth sensing provides an additional modality that complements the RGB information.

This project will use an RGBD camera to collect images with mirrors in them, compare the data captured in both modalities, and develop algorithms to integrate the information from both modalities to help mirror detection.

Some knowledge of computer vision and good programming skills are required for this project.