Data Elimination cum Interpolation for Imputation: A Robust Preprocessing Concept for Human Motion Data

The missing information from raw human motion data due to occlusion and hidden postures during motions has influenced the quality of input data. This study proposes a novel idea of preprocessing human motion data with data elimination cum interpolation for imputation to treat the missing information. The idea was implemented on three sets of public available motion data concerning jumping, walking and running activities obtained from YouTube (www.youtube.com). The video motions were transformed into numerical data with the aid of Photoshop tool in order to obtain the body segment markers in form of rotation angles and coordinates in the 2-dimensional format. The proposed approach was compared numerically to the conventional preprocessing approaches: data elimination, averaging, imputation to treat missing values. The efficiencies were confirmed by classification accuracies through BayesNet, Lazy Kstar, Decision table and Part method classifiers specifically chosen with the aid of WEKA tool. The findings demonstrated that data preprocessing using data elimination coupling with interpolation for imputation enhances the quality of human motion input data for better classification accuracy.