Affiliations: [a]
School of Artificial Intelligence, Hebei University of Technology, Tianjin, China
| [b] National Research Center for Rehabilitation Technical Aids, Beijing Key Laboratory of Rehabilitation Technical Aids for Old-Age Disability, Key Laboratory of Human Motion Analysis and Rehabilitation Technology of the Ministry of Civil Affairs, Beijing, China
Abstract: Sensor-based human activity recognition has a wide range of applications including caring the elderly, helping chronic patients, fitness, etc. As a new kind of single-layer feedforward network, extreme learning machine (ELM) has faster training speed and stronger generalization performance, which provides an effective technique for activity recognition. However, due to the random determination of input weights and hidden deviations, the ELM may converge to a local minimum in some cases. Therefore, in order to overcome the shortcomings of ELM and design a reliable and accurate recognition system, this paper proposes a multi-classifier recognition framework which utilizes extreme learning machines optimized by quantum-behaved particle swarm optimization (QPSO) as the base classifiers. The quantum-behaved particle swarm optimization was used to select the optimal parameters of base ELMs which are trained on different attribute characteristics. The proposed approach is assessed with two inertial sensor data sets. Comparative experiments with other optimization methods indicated that QPSO-ELM has better accuracy performance for inertial sensor-based human activityrecognition. The experiment showed that the proposed ensemble QPSO-ELM recognition method achieves an accuracy of 96.4% for recognizing six activities.
