Landmark-Based Methods for Temporal Alignment of Human Motions

Abstract

Abstract-Considerable current research is focusing on the subject of Wellbeing and aging, which covers a wide range of solutions to help encourage elderly people to engage on routine physical exercise. In particular, elderly communities in nursing homes are usually involved in activities on rehabilitation, daily exercises and health tracking. In order to implement an automated system for elderly exercise monitoring, human motion analysis should be efficiently performed in an affordable way, and delivered in a way that the users understand. In this paper, a general framework for comparison of fitness performances in the context of basic stand-up physical activities is presented. The Microsoft Kinect device is used for motion capture. A method for key body pose prediction on human activities based on multi-class C4.5, SVM, Naive Bayes and AdaBoost classifiers is first introduced, followed by a cluster analysis to refine the obtained results. The performances achieved with the different techniques and parameters are then analyzed and the best configuration compares favorably against the DTW and HACA algorithms.