Multi-model adaptation for thigh movement estimation using accelerometers

Abstract

Gait analysis plays an important role in healthcare and other applications. In the situation of ambulatory thigh movement estimation using accelerometers, the major challenges are non-linearity and uncertainty of thigh motion and variations of accelerometer measurement noise. In this study, the authors propose to use multiple motion models and noise models to meet these challenges. In order to adaptively select motion models and noise models to suit the thigh motion modes, feature vectors are derived from the acceleration signal in the wavelet domain for gait phases/modes detection. Based on the detection results, the right motion models and noise models are chosen, and an unscented Kalman filter is invoked to estimate the thigh movement using the chosen models. The experimental results have shown that the proposed method can estimate thigh movement accurately.