Recognizing Household Activities From Human Motion Data Using Active Learning and Feature SelectionBy Zhao, Liyue; Wang, Xi; Sukthankar, Gita; Technology and Disability, Vol. 22, No. 1-2, pp. 17-26
Publication Date: 2010
Article presents a method to recognize human household activities from motion data, using an active learning framework to improve sample efficiency and intelligent feature selection to reduce training time. The method improves classification of inertial measurement unit (IMU) data through intelligent feature selection. Signal data are converted into a set of motifs, approximately repeated symbolic subsequences, for each dimension of IMU data. These motifs leverage structure in the data and serve as the basis to generate a large candidate set of features from the multi-dimensional raw data. By measuring reductions in the conditional log-likelihood error of the training samples, features can be selected to train a conditional random field (CRF) classifier in recognizing human activities. Experiments with the techniques were carried out using the IMU unit of the publicly available Carnegie Mellon University Multi-Modal Activity Dataset (CMU-MMAC). The dataset consists of unscripted recipes performed by several participants in a kitchen. The proposed CRF method was compared against several baselines and sets of standard classifiers and was found to outperform them. The authors conclude that this method can effectively contribute to the development of home living assistance systems in the future.
Published by: IOS Press (Website:http://www.iospress.nl)
Association for the Advancement of Assistive Technology in Europe (AAATE) (Web Site: http://www.aaate.net )