Multisensor data fusion for human activities classification and fall detection

Abstract

Abstract-Significant research exists on the use of wearable sensors in the context of assisted living for activities recognition and fall detection, whereas radar sensors have been studied only recently in this domain. This paper approaches the performance limitation of using individual sensors, especially for classification of similar activities, by implementing information fusion of features extracted from experimental data collected by different sensors, namely a tri-axial accelerometer, a micro-Doppler radar, and a depth camera. Preliminary results confirm that combining information from heterogeneous sensors improves the overall performance of the system. The classification accuracy attained by means of this fusion approach improves by 11.2% compared to radar-only use, and by 16.9% compared to the accelerometer. Furthermore, adding features extracted from a RGB-D Kinect sensor, the overall classification accuracy increases up to 91.3%.

Magnetic and Radar Sensing for Multimodal Remote Health Monitoring

“…The data analyzed in this work were collected at the University of Glasgow with a group of 20 volunteers aged between 22 and 32 years. The ten activities are described in [20], and include: walking, walking while carrying an object with both hands, sitting on and standing up from a chair, bending to pick up a pen and to tie shoelaces, standing while drinking and answering a phone call, simulating a frontal fall onto a mat, and crouching to check under an imaginary bed and coming back up. A pictorial representation of these activities and their recording length are given in Fig.…”

“…In this paper, we expand our previous work in to consider a newer, larger database of sensors' signatures collected involving 20 volunteers aged 22-32 years. Although still limited, this appears to be in the top 3% in terms of number of subjects compared with some of the works on wearable for human motion analysis in the literature [13], [20]. Magnetic sensors have been often used jointly with accelerometer and gyroscopes for fall detection [7], [21], [22], or not considered in favour of using only data from the two aforementioned inertial units [23]- [26].…”

A Multisensory Approach for Remote Health Monitoring of Older People

“…1, involving 9 volunteers aged 23 to 31 years old. The activities were described in our previous work [18]. Three repetitions for each activity for each subject were recorded, generating a set of 270 sample measurements with simultaneous readings from the wearable sensor, the radar sensor, as well as a Microsoft Kinect recording for ground truth.…”

“…In this paper, we expand our preliminary work in [18] and present a detailed analysis of feature selection and information fusion methods when using simultaneous information from wearable sensors and radar sensor, for a new dataset comprised of new subjects. Inertial sensors are attractive for their compact form, low cost, relatively simple signal processing, and possibility of embedding into everyday objects such as phones or watches, which users may naturally take with them.…”

“…Furthermore, fusion approaches classically used for inertial sensors focus on signal level fusion, employing methods such as Vector Observation, Kalman filtering or other forms of signal filtering [13], whereas in our work we propose simpler fusion at feature and decision level, combined with effective feature selection approaches. Compared with the preliminary results presented in [18], this work considers additional features extracted from the wearable and radar sensor data, investigates three different methods to perform feature selection for each of these sensors, and presents different approaches for decision fusion. These include a voting-based system comprised of two SVM classifiers and two KNN classifiers, to exploit strengths from different classifiers and sensors in the decision process.…”