Knowledge discovery from 3D human motion streams through semantic dimensional reduction

Abstract

3D human motion capture is a form of multimedia data that is widely used in entertainment as well as medical fields (such as orthopedics, physical medicine, and rehabilitation where gait analysis is needed). These applications typically create large repositories of motion capture data and need efficient and accurate content-based retrieval techniques. 3D motion capture data is in the form of multidimensional time-series data. To reduce the dimensions of human motion data while maintaining semantically important features, we quantize human motion data by extracting spatio-temporal features through SVD and translate them onto a symbolic sequential representation through our proposed sGMMEM (semantic Gaussian Mixture Modeling with EM). In order to handle variations in motion capture data due to human body characteristics and speed of motion, we transform the semantically quantized values into a histogram representation. This representation is used as a signature for classification and similarity-based retrieval. We achieved good classification accuracies for “coarse” human motion categories (such as walking 92.85%, run 91.42%, and jump 94.11%) and even for subtle categories (such as dance 89.47%, laugh 83.33%, basketball signal 85.71%, golf putting 80.00%). Experiments also demonstrated that the proposed approach outperforms earlier techniques such as the wMSV (weighted Motion Singular Vector) approach and LB_Keogh method.