This repository contains the framework used for the master thesis with the same title. This project will also be published in the conference proceedings for the GSI 2019 conferencce.
The work is done is a collaboration with Professor Elena Celledoni and Dr. Nikolas Tapia.
This implementation is a continuation of work done by Markus Eslitzbichler while working as a PHD candidate at the Department of Mathematical Sciences, NTNU.
The animation folder contains two subfolders: src and db.
src/ contains all things animation related, that is Skeleton and Animation objects,
methods for parsing .asf/.amc-files, methods for creating animations and some
attempts at different frame interpolation.
db/ contains data and our database. To create the tables run:
sqlite3 <Name_of_db>.db < create_tables_sqlite3.sql
Download and unzip the mocap data, which can be obtained from http://mocap.cs.cmu.edu
create config-file: cp db_config_example.py db_config.py
and add the paths to your database and subject folder.
run:
python insert_data_db_sqllite3.py
to add data to database and download subject descriptions from mocap.cs.cmu.edu
animation_manager.py is an interface for fetching animations in
applications
The folder so3/ contains implementation our mathematical framework for SO3.
convert.py : convert animation to curce in SO3.
transformations.py log, exp, interpolate, SRVT and other transformations
applied to SO3 or curves in SO3.
curves.py: operations that take a curve, or multiple curves as
parameters. This includes distance, dynamic_distance, close, move_origin and
others. These are all written to be functional in style.
dynamic_distance.py: implementations off the the dynamic distance method
proposed by Bauer.
signature.py and log_signature.py: proposed metrics, calculated for
geodesic interpolation curves using the iisignature library.
the folders experiments/ and clustering/ contain different
applications of these methods.
Transformations applied to the group SE(3), the above mentioned framework could be applied to this group using a similar approach.