International Journal of Computer Theory and Engineering

Abstract—This paper proposes the use of interpolation methods rather that conventional learning algorithms such as Support Vector Machines (SVM) or Policy Learning by Weight Exploration with Return (POWER) for modelling human motion. The main aim was using a simpler model with less time and space complexity for later use in the recognition of certain actions. Three different polynomial interpolation methods, namely Lagrange, spline and cubic spline have been investigated. Parts of the dataset were used instead of the complete dataset using grouping techniques to reduce the training time. A non-parametric test known as Mc-Nemar's test was used to identify statistically significant performance differences between these methods. It was found that the cubic spline resulted in better accuracy.

Index Terms—Interpolations, dynamic movement primitives, learning algorithms, human motion.

Egemen Halici and Erkan Bostanci are with SAAT Lab., Computer Engineering Department, Ankara University, Golbasi, Ankara, Turkey (e-mail: egemenhalici@gmail.com, ebostanci@ankara.edu.tr).

[PDF]

Cite:Egemen Halici and Erkan Bostanci, "Evaluating the use of Interpolation Methods for Human Body Motion Modelling," International Journal of Computer Theory and Engineering vol. 10, no. 1, pp. 25-29, 2018.