Synopsis
The recent advent of commercial light field cameras, as well as depth cameras, opens up new opportunities for computer vision, with richer sensors. We are interested in the full stack of problems, from theoretical analysis of light field camera resolution, to shape recovery from multiple light field or RGBZ cues, to graphics applications, to higher-level vision applications in segmentation and recognition.
Project Details
Non-UCSD Vision People also Involved
Details
An important recent development is the commercial release of light field cameras from Lytro, Raytrix and Pelican. These cameras consider the whole light field rather than a single image. Since they keep the angular information, operations like refocusing after capture and changing one's perspective are also possible. However, light fields (and related depth cameras) have much broader implications in computer vision. The availability of multiple views and additional cues in a single shot passive capture enables new methods for shape recovery. Light field cameras also enable new approaches to higher-level computer vision tasks such as matting and compositing, or segmentation and recognition. Moreover, non-Lambertian reflectance and difficult photometry can be more easily handled. We seek a comprehensive end-to-end understanding from a core analysis of light field camera resolution to light field shape recovery to higher-level computer vision operations.