Activity Recognition Invariant to Sensor Orientation with Wearable Motion Sensors

Abstract

Abstract:Most activity recognition studies that employ wearable sensors assume that the sensors are attached at pre-determined positions and orientations that do not change over time. Since this is not the case in practice, it is of interest to develop wearable systems that operate invariantly to sensor position and orientation. We focus on invariance to sensor orientation and develop two alternative transformations to remove the effect of absolute sensor orientation from the raw sensor data. We test the proposed methodology in activity recognition with four state-of-the-art classifiers using five publicly available datasets containing various types of human activities acquired by different sensor configurations. While the ordinary activity recognition system cannot handle incorrectly oriented sensors, the proposed transformations allow the sensors to be worn at any orientation at a given position on the body, and achieve nearly the same activity recognition performance as the ordinary system for which the sensor units are not rotatable. The proposed techniques can be applied to existing wearable systems without much effort, by simply transforming the time-domain sensor data at the pre-processing stage.

Activity Recognition Invariant to Wearable Sensor Unit Orientation Using Differential Rotational Transformations Represented by Quaternions

“…In our earlier work [ 34 , 35 , 36 ], we have proposed two different approaches to transform and make time-domain sensor data invariant to the orientations at which the sensor units are fixed to the body. The first approach is a heuristic transformation where geometrical features invariant to the sensor unit orientation are extracted from the sensor data and used in the classification process [ 34 , 35 ], analogous to a method proposed by [ 37 ] for gait analysis. In the second approach, sensor sequences are represented with respect to three principal axes that are calculated using singular value decomposition (SVD) [ 34 , 36 ].…”

“…The first approach is a heuristic transformation where geometrical features invariant to the sensor unit orientation are extracted from the sensor data and used in the classification process [ 34 , 35 ], analogous to a method proposed by [ 37 ] for gait analysis. In the second approach, sensor sequences are represented with respect to three principal axes that are calculated using singular value decomposition (SVD) [ 34 , 36 ]. In both approaches, the transformed sequences are mathematically proven to be invariant to sensor unit orientations.…”

“…Unlike most of the other studies that investigate orientation invariance, the proposed heuristic and SVD-based methods are compared with the fixed orientation case and shown to decrease the accuracy by 15.5% and 7.6%, respectively, on average, over five publicly available datasets, four state-of-the-art classifiers, and two cross-validation techniques. It is also shown in [ 34 ] that randomly oriented sensor units degrade the accuracy by 21.2% when the untransformed sensor data are used for classification. In this article, as explained in Section 4.3 , we use a wider set of classifiers and leave-one-subject-out (L1O) cross-validation technique for better generalizability [ 38 ], in which case our newly proposed method achieves a noticeably higher accuracy than our previously proposed methods [ 34 , 35 , 36 ].…”

“…It is also shown in [ 34 ] that randomly oriented sensor units degrade the accuracy by 21.2% when the untransformed sensor data are used for classification. In this article, as explained in Section 4.3 , we use a wider set of classifiers and leave-one-subject-out (L1O) cross-validation technique for better generalizability [ 38 ], in which case our newly proposed method achieves a noticeably higher accuracy than our previously proposed methods [ 34 , 35 , 36 ].…”

IMU-Based Gait Recognition Using Convolutional Neural Networks and Multi-Sensor Fusion

“…The issue of ensuring orientation invariance in extracting gait features has been a matter of concern in many studies. In [ 20 ], the authors introduced an orientation invariant measure to alleviate the orientation dependency issue. In this work, we also employ an orientation invariant resulting measures of motion as the input signals.…”