Hi Sinan,

I recommend reading these papers about light field mapping / visual odometry

"Plenoptic Flow: Closed-Form Visual Odometry for Light Field Cameras", Donald G. Dansereau, Ian Mahon, Oscar Pizarro and Stefan B. Williams (2011)

"On the accuracy potential of focused plenoptic camera range determination in long distance operation", Hannes Sardemann, Hans-Gerd Maas (2016)

"Correlation-Based Depth Estimation with a Plenoptic Camera", William A. Roberts, and Brian S. Thurow (2016)

hope it helps.

Hi Sinan,

To estimate the depth in meters with correct precision using a Light Field Camera (in particular Lytro Illum product), a calibration stage is first required. Then the depth map needs to be estimated from the stereo focal stack. Finally, a rendering stage ends the pipeline.

Note : it is recommendded to work with focused plenoptic cameras to search for metric precision.

Check for the following usefull links:

https://hal.archives-ouvertes.fr/hal-01401501/document

http://www.grk1564.uni-siegen.de/sites/www.grk1564.uni-siegen.de/files/inm2013/plenoptic.pdf

http://www-personal.acfr.usyd.edu.au/ddan1654/PlenCal.pdf

https://pdfs.semanticscholar.org/2038/42dc302943f8658329de74572f0650e02c3b.pdf

http://www.isprs-ann-photogramm-remote-sens-spatial-inf-sci.net/III-3/449/2016/isprs-annals-III-3-449-2016.pdf

More general information may be found about plenoptic cameras by following these links:

http://persci.mit.edu/pub_pdfs/plenoptic.pdf

http://merkur229.inf.uni-konstanz.de/publications/GKWE15_book_chapter.pdf

http://f4k.dieei.unict.it/proceedings/ICIP2013/pdfs/0000001.pdf

http://www.tgeorgiev.net/rendering.pdf

Hoping that it will help you.

Hey Sinan,

I just came across your question, which also bothered me some time ago. Meanwhile, I have found two ways to estimate an object's distance by means of a light field camera and paraxial optics.

One attempt is to solve for the correspondence problem and find disparities from so-called sub-aperture images. Then it is possible to use an adaptation of the conventional stereo triangulation scheme, which I conceived for the Lytro-like plenoptic camera. Our most recent publication on this can be found here:

• Hahne et al. (Int J Comput. Vision, 2017)

https://doi.org/10.1007/s11263-017-1036-4

Alternatively, we can exploit a light field camera's capability to refocus. Here, a stack of images with varying focus is computed from a raw light field image where the sharpness/blur of each refocused photograph is analysed with a blur metric. The image exposing fewest blur for a particular object indicates the refocusing distance, which can be predicted as shown in:

• Hahne et al. (Optics Express, 2016)

https://doi.org/10.1364/OE.24.021521

More information on this can be found on my website:

• https://www.plenoptic.info

Soon, I will upload my codes there, which I have written in Python to process light field images as described above and get absolute distances to captured objects.

Hi Sinan,

I just stumbled across your question. In a previous project, I worked with the first Lytro camera model to determine the object distance. Since you already have managed to extract the disparity map you should be able to extract the depth information as long as you have done a calibration before. We also have written a paper, which could be helpful for your approach.

Feel free to contact me for any additional information on that topic.

Hope this helps.

Article Imagine All the Plants: Evaluation of a Light-Field Camera f...

Calibration via curve fitting (like in Robert's answer above) is a valid approach. However, zoom lens and optical focus settings just as temperature variations introduce degrees of freedom to the fitted curve that make calibration rather sophisticated.

In the design stage of a plenoptic camera, it is inevitable to understand the underlying ray optical model, a task which cannot be accomplished by the above method. For this purpose, I have released my distance estimator plenoptisign for the Lytro-like camera on GitHub:

https://github.com/hahnec/plenoptisign

which you can also run online:

http://www.plenoptic.info/pages/software.html

Edit:

This distance estimation can be used in conjunction with light field images rendered by plenopticam, an open-source application found at:

https://github.com/hahnec/plenopticam