Is is possible to perform 3D triangulation on two sets of 2D data without knowing camera parameters?

Hi all,

I am tracking the movement of an object inside a cube using two cameras - one above and one side-on.

I need to recreate the 3D path of this object in world coordinates however I can't use checkerboards etc to calibrate the cameras because the tracking data is being processed by another piece of software which imposes its own origin on the data.

Is it possible to use something similar to Matlab's triangulate function but without access to camera matrices? perhaps using the corners of my cube instead somehow?

Thanks for any help,

Rod.

[Walter Roberson]

Is the cube fixed position in the image? And fixed size? Could you post a sample?

I think it might be do-able provided you know the size of the cube. If it is cuboid rather than a true cube then it would get much more complicated in the situation where it also appears in different orientations (requiring that it be deduced which of the edges of the cube were being examined.)

[Right Grievous]

Hi Walter,

Sorry for the slow reply, I forgot that I don't get email alerts for comments.

I have attached some sample data (the outside shape of the cube which I am using to test scripts) and a script to load the data and plot it.

I don't know the size of the cube off hand, but I can find that out. It is a perfect cube in reality, it's just distorted in the data due to perspective issues.

I'm not sure what you mean by different orientations... the cameras should never move, there will always be one above and one to the side. The cube should always be in the same place/orientation as well, within a few cm.

Thank you for any help,

Rod.

[Thomas Koelen]

May I ask why you are using a different software packing to do the tracking?

[Right Grievous]

Well we are measuring several features of the object (temp, electrical conductance etc) which are processed in a separate piece of hardware, which requires its own software.

To keep all of the data synchronized (we need micro second accuracy) we have to really use this software for tracking too.

Besides, I have other problems - the cameras we are using are great for tracking in IR but they don't capture very high quality images which makes camera calibration difficult. Furthermore, these cameras use a BNC connection, so to take stills with them I have to run the signal through a graphics adapter which also no doubt interferes with the images in such a way as to make my tracked coordinates no longer align with the captured stills... it's all a bit of a mess.

I do know where the corners of the cube are, and that the cube is a nice square cube, so I'm hoping to sort out the data using this as a reference.