More Projected Geometry

Projected geometry has improved! Surface extraction for a cistern has made it through the pipeline and (as shown below) is now visualized. At the moment, we have a simple keyhole cistern... one of the most common outline shapes for cisterns. It is interesting to see how the sonar scans picked up other objects at the level of the scan that have been interpolated into vertical barriers (seen as pillars).

In addition to surface extraction, I now have multiple projections that can be merged together as a mosaic on the wall of the cistern. Below you can see the mosaic merged together in Microsoft Word on the left. On the right, the same mosaic is replicated in the 3D environment on a wall of the keyhole cistern. The images are actually taken from the ISIS boat trip. On the ISIS, we recorded the ROV's video of the shipwreck. From the video, I took screenshots that represent sequential frames of different features of the shipwreck. With a little patience, I put the images together (as accurately as I could) to represent the shipwreck.

This process will be the same for combining images from the all of the surveyed sites with their corresponding geometry. The mosaic process of projective texturing still has a lot of work to help streamline the process more. Just for the record... this is all REALLY cool :P

Projective Texturing of Cisterns

For the final presentation of the cistern exploration project, all of the video and images will be mapped to the geometry generated from the sonar data. The screenshot below is a mock world generated to serve as example data extracted from sonar. Using mathematical calculations, the images are loaded into a program and projected onto the geometry (just like you would think a projector to work) to make the world look more realistic and similar to what it actually looked like from the perspective of the robot. There is currently only one projection of one image taken of a cistern a couple of years ago.

At this point, there is certainly more work to be done to merge our actual data and images together. The previous plan was to automate the video frame extraction with the location and orientation of the robot and project it. After completing many projects, we realized that it would be near impossible to complete this automatically and accurately.

Currently, my plan is to (once the geometry is loaded into the program) load images that correspond to the cistern and scan through them until one matches the current orientation within the world. At this point, the image projection will be saved, and a new image can be loaded to project and save to overlay or stitch with previous projections. This method will be an easy way to accurately bring multiple images together and place them in the right place in the geometry to make the world realistic.