**Yesterday:**Constructed physical simulation of balls bouncing on a table, and used application to see about object segmentation. Sometimes works. (video, fast) Many times has erroneous identification. Occassionally marks all vertices in a single group, while other times it marks all vertices in separate groups (which is definitely wrong, given a table should be static). Need to check for bugs:

- Are the vertices getting reliably re-ordered?
- What do the block adjacencies look like when finding all vertices the same? When finding all vertices different?
- Are there other bugs in the program?

Provided example for 2, 10 and 49 balls dropping. Worsens as more balls are introduced. Tweaked application so that only current files needed are loaded into memory.

**Today:** Look into bugs from yesterday. Chat with supervisors about how it looks. Create series of synthetic double pendulum example. Implement transformation-based validation (ie. modification of original step 2 of process), and process for reliance on prior frames. Generate a file-loading for test cases so there is consistency in process validation.

Write up (mathematical foundation, computational/theoretical implementations) for future writing/papers:

- distance-based segmentation
- block-based segmentation/checking

Consider laying out how I'd like the thesis to look and how I'd like it to describe the various implementations and stages along the way.

**Roadblocks:** Not being overwhelmed, being realistic, and taking it baby-step by baby-step.

**Where Does this Fit In:** Having 3D synthetic examples to work with is important in validating the developed reconstruction process. Developing good visualizations - as well as a mathematical discussion of the process - along the way helps demonstrate proof-of-concepts, provides affirmation of skill sets for resource/backing requests, and justifies the direction to be taken.