Virtual Reality plays the leading role on the new media revolution with Light Field reconstruction as a common technique for content generation.High industrial interests make this technique hard to understand and implement, I will present a novel method to reconstruct the depth of objects in images.
Light Field reconstruction is a relatively new technique for 3D reconstruction and rendering, that can be used to compute the depth of objects from a particular scene (physical subspace) given a set of images from different viewpoints. This technique has gained a lot of interest in the last ten years due the applications to Virtual and Augmented Reality; reflected most recently on the acquisition of the company Lytro (Light Field Camera Developer) by Google on March of this year (https://www.theverge.com/2018/3/21/17148622/google-lytro-acquisition-light-field), and the recent advances presented also by Google (https://www.blog.google/products/google-vr/experimenting-light-fields/) and Magic Leap (https://www.wired.com/story/magic-leap-lightwear-headset-hardware/) for Light Field Applications on the new VR media revolution.
Most of the Light Field reconstruction methods are computationally expensive, and depending on very large data sets of high resolution images. In addition with this issue, the existing research on this field lacks of clarity and reproducibility, this mainly caused by the high industrial interests behind.
Inspired by the search of interesting applications of the julia package Shearlab.jl (https://github.com/arsenal9971/Shearlab.jl, developed by me and presented at the last JuliaCon and PyData Berlin) and the open source spirit, I developed a julia library for Light Field reconstruction that optimizes the computation by using the Shearlet transform to represent sparsely the Light Field, called LightFields.jl (https://github.com/arsenal9971/LightFields.jl).
This software is not just fully open source but presents clear improvements in the processing time of Light Fields in comparison with most of the modern industrial solutions. In addition to this I developed a open hardware Light Field camera with a Raspberry Pi board that uses the library, and cost more than five times less than the existing commercial Light Field cameras. I would like to present both software and hardware at the PyData Berlin 2018.