School of Control Science and Engineering, Dalian University of Technology, Dalian, Liaoning, People's Republic of China.
Physiol Meas. 2011 Mar;32(3):347-58. doi: 10.1088/0967-3334/32/3/006. Epub 2011 Feb 18.
Human activity recognition (HAR) by using wearable accelerometers has gained significant interest in recent years in a range of healthcare areas, including inferring metabolic energy expenditure, predicting falls, measuring gait parameters and monitoring daily activities. The implementation of HAR relies heavily on the correctness of sensor fixation. The installation errors of wearable accelerometers may dramatically decrease the accuracy of HAR. In this paper, a method is proposed to improve the robustness of HAR to the installation errors of accelerometers. The method first calculates a transformation matrix by using Gram-Schmidt orthonormalization in order to eliminate the sensor's orientation error and then employs a low-pass filter with a cut-off frequency of 10 Hz to eliminate the main effect of the sensor's misplacement. The experimental results showed that the proposed method obtained a satisfactory performance for HAR. The average accuracy rate from ten subjects was 95.1% when there were no installation errors, and was 91.9% when installation errors were involved in wearable accelerometers.
近年来，使用可穿戴加速度计进行人体活动识别 (HAR) 在医疗保健领域引起了广泛关注，包括推断代谢能量消耗、预测跌倒、测量步态参数和监测日常活动。HAR 的实施在很大程度上依赖于传感器固定的正确性。可穿戴加速度计的安装误差可能会极大地降低 HAR 的准确性。本文提出了一种方法，以提高 HAR 对加速度计安装误差的鲁棒性。该方法首先使用 Gram-Schmidt 正交归一化来计算变换矩阵，以消除传感器的方向误差，然后使用截止频率为 10 Hz 的低通滤波器来消除传感器错位的主要影响。实验结果表明，所提出的方法在 HAR 中取得了令人满意的性能。在没有安装误差的情况下，来自十个受试者的平均准确率为 95.1%，而在涉及可穿戴加速度计的安装误差的情况下，准确率为 91.9%。
