Jan Sedmidubsky October 3, 2013Retrieving Similar Movements in Motion Capture Data
Jan Sedmidubsky
Jakub Valcik
Retrieving Similar Movements in
Motion Capture Data
Faculty of Informatics
Masaryk University
Brno, Czech Republic
1/7
Jan Sedmidubsky
• Motion Capture Data ~ Human Motion Data
– Sequence of poses of 3D joint coordinates
• Capturing devices:
– Tracking markers attached to a human body (e.g., Vicon)
– Systems of synchronized cameras (e.g., Microsoft Kinect)
– Estimating 3D joint coordinates from ordinary videos
October 3, 2013Retrieving Similar Movements in Motion Capture Data
Introduction
2/7
Jan Sedmidubsky October 3, 2013Retrieving Similar Movements in Motion Capture Data
Introduction
• Analysis of recorded motion data in various areas:
– Health care – success of rehabilitative treatments
– Sports – performance aspect comparison
– Security – person identification, event detection
– Computer animation – realistic motion synthesis
• Challenge – increase findability of recorded motions
– Annotations – limited to given motion classes
– Content-based retrieval – requiring a query example
3/7
Similar?
Jan Sedmidubsky October 3, 2013Retrieving Similar Movements in Motion Capture Data
Content-based (Sub-)motion Retrieval
• Content-based retrieval:
– Search for motions in a database that are similar to a
query motion example
– Approaches:
• Sequence-based approach – searches for entire motions only
• Subsequence-based approach – searches for all possible submotions
within recorded database motions
– Components:
• Similarity model
– Motion features (descriptors)
– Motion similarity function considering temporal variances
• Indexing & searching
4/7
Jan Sedmidubsky
• Similarity model:
– Motion features – joint-angle rotations
• Each pose = 28-D vector of angles of joints
• Individual poses compared by the L1 metric
– Motion similarity function
• Average distance between key poses based on the L1 metric
• Indexing & searching:
– Indexing all motion poses (28-D vectors) by the L1 metric
• Any metric-based structure (e.g., Metric-Index [Novak, 2011])
– A specialized key-pose retrieval algorithm
• Sedmidubsky, J., Valcik, J., and Zezula P. A Key-Pose Similarity Algorithm for
Motion Data Retrieval. In 12th International Conference on Advanced Concepts for
Intelligent Vision Systems (ACIVS 2013). Springer, 2013.
October 3, 2013Retrieving Similar Movements in Motion Capture Data
Our Sub-motion Retrieval Approach
5/7
Jan Sedmidubsky October 3, 2013Retrieving Similar Movements in Motion Capture Data
Our Sub-motion Retrieval Approach
Online Demonstration
• Online demo: http://disa.fi.muni.cz/motion-retrieval/
• HDM05 motion database [Muller, 2005]:
– 102 motions of 491,847 frames (poses) ~ 68 minutes
6/7
Jan Sedmidubsky October 3, 2013Retrieving Similar Movements in Motion Capture Data
Conclusions & Future Work
• Conclusions:
– Content-based subsequence search in motion data
– Online web application for sub-motion retrieval
• Future research directions:
– Improving retrieval efficiency to achieve sublinear search
costs with respect to the length of database motions
– Developing new similarity models to achieve better
retrieval effectiveness
7/7
Jan Sedmidubsky October 3, 2013Retrieving Similar Movements in Motion Capture Data
Questions?
Thank you for your attention.
Try our online demo:
http://disa.fi.muni.cz/motion-retrieval/