Human Activity Recognition (HAR) is an important application of smart wearable/mobile systems for many human-centric problems such as healthcare. The multi-sensor synchronous measurement has shown better performance for HAR than a single sensor. However, the multi-sensor setting increases the costs of data transmission, computation and energy. Therefore, the efficient sensor selection to balance recognition accuracy and sensor cost is the critical challenge. In this paper, we propose an Instance-wise Dynamic Sensor Selection (IDSS) method for HAR. Firstly, we formalize this problem as minimizing both activity classification loss and sensor number by dynamically selecting a sparse subset for each instance. Then, IDSS solves the above minimization problem via Markov Decision Process whose policy for sensor selection is learned by exploiting the instance-wise states using Imitation Learning. In order to optimize the parameters of the activity classification model and the sensor selection policy, an algorithm named Mutual DAgger is proposed to alternatively enhance their learning process. To evaluate the performance of IDSS, we conduct experiments on three real-world HAR datasets. The experimental results show that IDSS can effectively reduce the overall sensor number without losing accuracy and outperforms the state-of-the-art methods regarding the combined measurement of accuracy and sensor number.

Surface electromyography (sEMG) array based gesture recognition, which is widely-used, could provide natural surfaces for human-computer interaction. Currently, most existing gesture recognition methods with sEMG array only work with the fixed and pre-defined electrodes configuration. However, changes in the number of electrodes (i.e., increment or decrement) is common in real scenarios due to the variability of physiological electrodes. In this paper, we study this challenging problem and propose a random forest based ensemble learning method, namely feature incremental and decremental ensemble learning (FIDE). FIDE is able to support continuous changes in the number of electrodes by dynamically maintaining the matrix sketches of every sEMG electrode and spatial structure of sEMG array. To evaluate the performance of FIDE, we conduct extensive experiments on three benchmark datasets, including NinaPro, CSL-hdemg, and CapgMyo. Experimental results demonstrate that FIDE outperforms other state-of-the-art methods and has the potential to adapt to the evolution of electrodes in the changing environments. Moreover, based on FIDE, we implement a multi clients/server collaboration system, namely McS, to support feature adaption in real-world environment. By collecting sEMG using two clients (smartphone and personal computer) and adaptively recognizing gestures in the cloud server, FIDE significantly improves the gesture recognition accuracy in electrode increment and decrement circumstances.

No abstract