A Comparison Method To Evaluate Motion Splicing Techniques

The development of methods for the re-use and modification of motion data is an activearea of research. Among these methods, we find automatic techniques to generate a spliced motion by combining different actions recorded in separate sessions. Motion splicing allows capturing motion independently and later combining them to create a new natural looking motion. Even though there have been a lot of research on motion editing techniques, less focus has been given to evaluate these edited motions. In this thesis, we present a novel methodology to quantitatively evaluate the synthesized motion generated by different motion editing techniques. We implemented three splicing algorithms to perform a comparison study based on our evaluation methodology. The splicing algorithms considered are spatial body alignment, segmentation-based method, and naïve DOF replacement. We use 39 sets of motions in the Human Motion Database specifically collected for testing splicing. The motion sets consist of two actions performed individually and then in combination to each other. This data served as the ground truth in our comparison. The motions synthesized using the above splicing techniques were then evaluated using our quantitative evaluation method. The spatial body alignment performs best since it accounts for the correlation between joints. Due to poor segmentation, the segmentation-based method generated too much jerkiness in the synthesized motion which was demonstrated with a poor result in our evaluation