Embedded Files
This is going to be an attempt to come up with the methodology for making this work. If you are wanting to try this out, please wait until all of the parts are gathered and it is tested, or go ahead and let me know what works best for you.
There are two cameras online that meet the sub $100.00, global shutter, and compact enough work with a 3D printed holder inside the hyperbolic lens. The cameras are IDS uEye UI-1226LE-C-HQ and KYT-U030-GS01. In the end I decided to go with the monochrome UI-1226 camera (there was a used one on ebay for about $30.00).
I have purchased different types of lenses in order to test out the best for the best image of the environment. To make this work, a new camera-lens mount (mount to hold the camera to the mirror) will be designed and used. The STL file will be added to thingiverse when it is done and tested.
I have been thinking a lot about how to get this working. These are the two basic theories on how to make this work:
- Use an omnidirectional camera calibration toolkit and use the output to use with LSD-SLAM.
- Dewarp the imaged frame by frame, cut the image into smaller sections, and then reference the image feeds as virtual cameras.
I want to cover a few things before we continue. I believe I have erroneously called the rounded omnidirectional mirror a parabolic mirror--which are defined as hyperbolic mirrors or catadioptric system by scholarly articles. This mirror placed in front of a camera with a lens makes it a catadioptric camera system. Another term you may want to know if you need to search for similar setups is ODVS--Omni-Directional Vision Systems. With this information, you should be able to find some good documentation written by brilliant people who could change the world--if they released source code.... With any luck you may even find the document that leads me to believe that LSD-SLAM can be used with such a catadioptric system: https://www.gc.cuny.edu/CUNY_GC/media/Computer-Science/Student%20Presentations/Carlos%20Jaramillo/Second_Exam_Survey_Carlos_Jaramillo_5_27_2016.pdf
I have ran into more many issues with the camera, lenses, ideas for the final design of the omnidirectional vision system, and my limited competence in each of these subjects. This has slowed my ability to throw the camera together in a seemingly magical way that so many internet people are able to do. Behold the problems:
Above is the half-sectional view of the newest bad ideas, or poor designs in camera-lens adapters. The camera and lenses came in faster than anticipated, so I started CADing fast and poorly. The problem with this design is that focus is limited due to the distance from the mirror. This mistake was make because the camera that this was tested on had a zoom and a focus, the new camera has neither. The lens was then adjusted to help correct for this distance issue but there were more problems.
The camera purchased from Ebay was not stock. The lens mount was different from all others I have seen on the web. The screw placement was on the four corners of the CMOS rather than two middle outsides of typical lens holders. The lens connection was a standard M12 type. The lens mount was also longer than typical, making it hard to get the lens focused and/or use other lenses.
Because the standard lens mounts would not fit this camera, part of the mount had to be removed with a hacksaw. By removing a portion of the lens mount all of the lenses purchased worked with the camera.
A new camera-mirror mount was designed to allow for a better picture of the mirror:
The first design (yellow) was made to work with the camera being place facing up from the middle hole. This new design is built to have the camera placed facing down-ward through the rectangle hole, and then turned upside down and placed inside the mirror. This results in the need for a 90 degree micro usb cable.
Hopefully one day we will get it calibrated, but not today.
This page is not finished.
10/15/17
Google Sites
Report abuse