Mark M. Miller

Masters Project

For my MSc project I optimised and analysed the effectiveness of different hardware parallelism on a motion retargetting algorithm, using OpenCL. The goal of this project was to create a system that allowed rendering of hundreds of characters in real-time whilst also adapting to deformities in the terrain. My system made effective use of the GPU to render and retarget hundreds of characters in real-time in comparison with just a few characters in the original system.

The motion adaptation method was based on Relationship Descriptors which allows retargetting of character joints based on descriptors embedded in the geometry near the character. The descriptors move with the deformed terrain and the joints are retargetted based on these descriptors to create the adapted motion.

This project was submitted and accepted as the basis for a paper at MIG 2015