-1

I have a collection of camera images taken by 4 calibrated cameras mounted on a mobile robot moving in a static outdoor environment. In addition, I have information from a farely accurate OxTS RT3000 Inertial Navigation System (IMU + GPS).

I would like to combine these images to form a 3d model (point cloud) of the static environment. I know there are many Structure from Motion applications, but I would like to find some software/library that is able to make use of the odometry and calibration, at least as an initialization, and to produce a dense point cloud. (All of this is for offline recordings.)

Any suggestions?

Daniel P
  • 199
  • 2
  • 15

1 Answers1

0

Agisoft Photoscan does what you want. From their manual:

PhotoScan supports import of external and internal camera orientation parameters. Thus, if precise camera data is available for the project, it is possible to load them into PhotoScan along with the photos, to be used as initial information for 3D reconstruction job.

Taken from page 21 of http://www.agisoft.com/pdf/photoscan-pro_1_4_en.pdf

You may have to do some wrangling of data to get it into a supported format, but it's certainly possible. You'll probably want to use the local coordinate output of the OxTS IMU to go from lat/lon/alt to XYZ and save yourself the conversion. Be careful also to correct for the extrinsic parameters - the rotations and translations between the IMU navigation frame and the cameras.

DoMakeSayThink
  • 155
  • 6
  • Thanks, that looks very good (although a little pricey)! I am correcting for the extrinsic parameters, even though it's not perfect, as I only measure the position and rotation of the IMUs casing. I do not have an automatic method to do the relative calibration between the cameras and the IMU. – Daniel P Nov 01 '18 at 21:25
  • Yeah, it is expensive software but near enough standard for professional photogrammetry work. I'm not sure how well it deals with the errors that would be introduced from imprecise extrinsics, but you'd expect bundle adjustment to account for it. – DoMakeSayThink Nov 02 '18 at 14:34