Undergrad Research Project - Using polarization cues to reconstruct three dimensional shape

Spring 2017

Zecheng He
Aswin Sankaranarayanan
Project description

Last semester my partner worked on using polarization cues to enhance depth information. She managed to develop an algorithm based on existing papers which can be used to regenerate surface. However, there are several problems with this algorithm. First of all, the normal information acquired from polarization is prone to have noise at regions perpendicular to the camera. In addition, the algorithm used to combine depth and normal information is not very effective. Therefore, we would like to spend this semester seeking solutions for these problems.

Me and my group would like to do the following to approach these problems: 1) We will study existing papers. In addition to polarization with single cameras, photometric stereo and polarization from multiple viewpoints are both approaches people have taken to tackle this problem. 2) Based on the knowledge we gain from the papers, we will develop a new method to acquire data, and new algorithm to process acquired data. 3) We will build a camera system to experimentally verify our method.

If we manage to improve from last semester's result, the noise from polarization normals can likely be resolved, thus the development of new algorithm will be less challenging.

Return to project list