Improving Our Volumetric Pipeline

As you can see from my lasts couple of posts, I have been working a lot on volumetric capture solutions here at the lab. Recently, Zhenyi and I have been trying to streamline the pipeline we have constructed to record a person from different cameras, and merge the point cloud data from these cameras easily. I have worked on incorporating a final registration pass over the point clouds, in order to seamlessly combine them. Zhenyi has also streamlined our ability to calibrate the cameras.

Without having to take photos manually, we now have a script that captures a stream of chessboard images as we rotate it in realtime.

Example of Chessboard Calibration
Where the calibration has detected our cameras to be.

If we don’t get completely perfect results, even with calibration, we use an algorithm called SparseICP as a post-processing pass. This gives us some pretty accurate results.

Point cloud with registration pass
Combined point cloud with only calibration

Tune in for next time to see my progress : )

Leave a Comment

Your email address will not be published.