View Full Version : Mounting 2 Cameras for Stereo Vision

05-22-2014, 10:22 PM
Hi, I'm new to robotics - but fairly excited to start - just wanna get an idea if I can mount 2 of the MS Cameras on a phantom x turret.

I am not sure what kind of brackets come with it, or if I need additional equipment for that kind of setup. Prefer not to have to resort to duct tapes if I at all possible :P

http://www.trossenrobotics.com/p/Microsoft-LifeCam-Cinema-Webcam-HD-robot-camera.aspx (X2)


05-23-2014, 12:03 PM
I believe you need to make some additional kind of bracket for that, but I don't have the turret, so I'm not sure.

I use two MS cameras for my rover, and use OpenCV to calibrate them for stereo. I made the bracket out of 1 inch square aluminum tubing.


05-23-2014, 01:18 PM
Urk... I rather not be doing any drilling in my apartment... besides I just want to code and spent as little time on hardware as possible.

05-23-2014, 03:42 PM
Drilling 3mm holes in plastic with a cordless drill is totally apartment compatible :-)
Although I imagine some JB Weld or other epoxy-like adhesive might also work.
Fasten the stand-offs in the right pattern on the top of the quad walker, and screw the top of the turret into those stand-offs.

05-25-2014, 04:15 AM
For multi-camera vision you get the best results if you have a full definition of where all the cameras are. With a bracket, you'll be much more positive as to the exact location of the cameras.

05-25-2014, 02:32 PM
The main requirement is that the cameras don't move after calibration. As long as you can rigidly fix the cameras, then a bracket, some InstalMorph, or superglue can all achieve the same rigid fixing of the cameras, with varying degree of aesthetics.
Once the cameras are fixed, you need to run calibration of some sort; typically print out a chessboard pattern and take a bunch of pictures and calculate the camera distortion and projection based on detecting that pattern. OpenCV has functions for this.
Once calibrated, you can rectify left/right input pictures so the same scanlines represent the same horizontal lines in the world, which simplifies the work of trying to figure out matches vs disparities between the cameras.