Detecting postural transitions: A robust wavelet-based approach

Abstract

The ability to perform postural transitions such as sit-to-stand is an accepted metric for functional independence. The number of transitions performed in real-life situations provides clinically useful information for individuals recovering from lower extremity injury or surgery. Performance deficits during these transitions are well correlated to negative outcomes in numerous populations. Thus, continuous monitoring and detection of transitions in individuals outside of the clinical setting may provide important, clinically relevant information regarding the progression of physical impairments. In this paper, we propose a new inertial-sensor based approach to detecting transitions utilizing the wavelet transform. This approach performs robustly in both supervised laboratory settings, and in ambient settings. We evaluate the performance of our algorithm on a data set including 334 in-laboratory and 20 in-home postural transitions from individuals with and without motor impairments.

A Systematic Review of Time Series Classification Techniques Used in Biomedical Applications

“…The data were collected from 16 healthy subjects, and non-linear features were extracted, selected, and fed into a Random Forest Classifier, thereby achieving an accuracy score of 0.99, F1-score of 0.99, and Cohen’s Kappa of 0.985. Among the papers that used small datasets, (discrete or continuous) wavelet transform—as a processing method or feature extraction technique—was very commonly used and very effective, such as for the paper presented by Hemmati et al [ 75 ] (mentioned above), which used data from only 12 subjects. Statistical modeling methods are also very effective as classifiers, such as the paper presented by She et al [ 33 ] (mentioned above), which used data from only 20 subjects.…”

“…Inertial Measurement Unit (IMU): Hemmati et al [ 75 ] presented a wavelet-based algorithm to detect postural transitions. The inertial signal was decomposed using a 4 th -order Daubechies Wavelet Transform and the classifier uses subject-specific fixed thresholds (curve length and area under the curve) to achieve an accuracy as high as 0.96.…”

A Secure Voice Signature Based Lightweight Authentication Approach for Remote Access Control

“…The number of transitions performed in real life situations provides useful clinical information for an individual recovering from lower extremity injury or surgery. Consequent to this, Sadra and Eric [5] proposed a new inertialsensor based approach to detect transitions using wavelet transform. Their approach is robust for supervised laboratory and ambient settings.…”