Natalie Baddour

Feature Selection for Wearable Smartphone- Based Human Activity Recognition with Able bodied, Elderly, and Stroke Patients

Human activity recognition (HAR), using wearable sensors, is a growing area with the potential to provide valuable information on patient mobility to rehabilitation specialists. Smartphones with accelerometer and gyroscope sensors are a convenient, minimally invasive, and low cost approach for mobility monitoring. HAR systems typically pre-process raw signals, segment the signals, and then extract features to be used in a classifier. Feature selection is a crucial step in the process to reduce potentially large data dimensionality and provide viable parameters to enable activity classification. Most HAR systems are customized to an individual research group, including a unique data set, classes, algorithms, and signal features. These data sets are obtained predominantly from able-bodied participants. In this paper, smartphone accelerometer and gyroscope sensor data were collected from populations that can benefit from human activity recognition: able-bodied, elderly, and stroke patients. Data from a consecutive sequence of 41 mobility tasks (18 different tasks) were collected for a total of 44 participants. Seventy-six signal features were calculated and subsets of these features were selected using three filter-based, classifier-independent, feature selection methods (Relief-F, Correlation-based Feature Selection, Fast Correlation Based Filter). The feature subsets were then evaluated using three generic classifiers (Naïve Bayes, Support Vector Machine, j48 Decision Tree). Common features were identified for all three populations, although the stroke population subset had some differences from both able-bodied and elderly sets. Evaluation with the three classifiers showed that the feature subsets produced similar or better accuracies than classification with the entire feature set. Therefore, since these feature subsets are classifier-independent, they should be useful for developing and improving HAR systems across and within populations.

Wearable Mobility Monitoring Using a Multimedia Smartphone Platform

Understanding mobility is important for effective clinical decision making in the area of physical rehabilitation. Ideally, a persons mobility profile in a nonclinical setting, such as the home or community, could be obtained. This profile would include the environment and context in which the mobility takes place. This paper introduces a novel wearable mobility monitoring system (WMMS) for an objective ubiquitous measurement of mobility. This prototype WMMS was created using a smartphone-based approach that allowed for an all-in-one WMMS. The wearable system is freely worn on a persons belt, such as a normal phone. The WMMS was designed to monitor a users mobility state and to take a photograph when a change of state was detected. These photographs were used to identify the context of mobility events (i.e., using an elevator, walking up/down stairs, and type of walking surface). Validation of the proposed WMMS was performed with five able-bodied subjects performing a structured sequence of mobility tasks. System performance was evaluated by its ability to detect changes of state and the ability to identify context from the photographs. The WMMS demonstrated good potential for community mobility monitoring.

Operational and convolution properties of two-dimensional Fourier transforms in polar coordinates

For functions that are best described in terms of polar coordinates, the two-dimensional Fourier transform can be written in terms of polar coordinates as a combination of Hankel transforms and Fourier series-even if the function does not possess circular symmetry. However, to be as useful as its Cartesian counterpart, a polar version of the Fourier operational toolset is required for the standard operations of shift, multiplication, convolution, etc. This paper derives the requisite polar version of the standard Fourier operations. In particular, convolution-two dimensional, circular, and radial one dimensional-is discussed in detail. It is shown that standard multiplication/convolution rules do apply as long as the correct definition of convolution is applied.

Correcting Smartphone orientation for accelerometer-based analysis

A method was developed for rotating a Smartphone accelerometer coordinate system from an offset to a predetermined three-dimensional position to improve accelerometer-based activity identification. A quaternion-based rotation matrix was constructed from an axis-angle pair, produced via algebraic manipulations of the gravity acceleration components in the devices body-fixed frame of reference with the desired position of the vector. The rotation matrix is constructed during quiet standing and then applied to all subsequent accelerometer readings thereafter, transforming their values in this new fixed frame. This method provides a consistent accelerometer orientation between people, thereby reducing Smartphone orientation variability that can adversely affect activity classification algorithms.

Mobility change-of-state detection using a smartphone-based approach

Understanding the mobility of people with physical disabilities is important for rehabilitation decision making. This paper presents a smartphone-based approach to mobility monitoring. The BlackBerry-based system is clipped to the persons belt. This approach uses an accelerometer signal to identify changes-of-state caused by starting/stopping and postural change. Our finding suggests that a smartphone integrated with an accelerometer could detect changes from static or dynamic movement (i.e., starting to walk, standing still, slowing down), which compares favorably with previous studies using body-fixed accelerometers. This approach is part of the larger framework of Wearable Mobility Monitoring Systems (WMMS).

Signal processing of heart signals for the quantification of non-deterministic events

BackgroundHeart signals represent an important way to evaluate cardiovascular function and often what is desired is to quantify the level of some signal of interest against the louder backdrop of the beating of the heart itself. An example of this type of application is the quantification of cavitation in mechanical heart valve patients.MethodsAn algorithm is presented for the quantification of high-frequency, non-deterministic events such as cavitation from recorded signals. A closed-form mathematical analysis of the algorithm investigates its capabilities. The algorithm is implemented on real heart signals to investigate usability and implementation issues. Improvements are suggested to the base algorithm including aligning heart sounds, and the implementation of the Short-Time Fourier Transform to study the time evolution of the energy in the signal.ResultsThe improvements result in better heart beat alignment and better detection and measurement of the random events in the heart signals, so that they may provide a method to quantify nondeterministic events in heart signals. The use of the Short-Time Fourier Transform allows the examination of the random events in both time and frequency allowing for further investigation and interpretation of the signal.ConclusionsThe presented algorithm does allow for the quantification of nondeterministic events but proper care in signal acquisition and processing must be taken to obtain meaningful results.

A revisit of spinning disk models. Part I: derivation of equations of motion

Abstract Previous models of spinning disks have focused on modelling the disk as a spinning membrane. The effect of bending stiffness was then incorporated by adding the appropriate term to the previously derived spinning membrane equation. A pure spinning plate model does not exist in the literature. Furthermore, in both existing linear and nonlinear models of spinning disks, the in-plane inertia and rotary inertia of the disk have been ignored. This paper revisits the derivation of the equations of motion of a spinning plate. The derivation focuses on the use of Hamiltons principle with linear Kirchhoff and nonlinear von Karman strain expressions. In-plane and rotary inertias of the plate are automatically taken into account. The use of Hamiltons principle guarantees the correct derivation of the corresponding boundary conditions. The resulting equations and boundary conditions are discussed.

Change-of-state determination to recognize mobility activities using a BlackBerry smartphone

A Wearable Mobility Monitoring System (WMMS) can be a useful tool for rehabilitation decision-making. This paper presents preliminary design and evaluation of a WMMS proof-of-concept system. Software was developed for the BlackBerry 9550, using the integrated three axes accelerometer, GPS, video camera, and timer to identify mobility changes-of-state (CoS) between static activities, walking-related activities, taking an elevator, bathroom activities, working in the kitchen, and meal preparation (five able-bodied subjects). This pilot project provides insight into new algorithms and features that recognize CoS and activities in real-time. Following features extraction from the sensor data, two decision trees were used to distinguish the CoS and activities. Real-time CoS identification triggered BlackBerry video recording for improved mobility context analysis during post-processing.

A smartphone approach for the 2 and 6-minute walk test

The 2 and 6-minute walk tests (2-6MWT) are used by rehabilitation professionals as a measure of exercise capacity. Our research has produced a new 2-6MWT BlackBerry smartphone application (app) that can be used to run the 2-6MWT and also provide new information about how the person moves during the test. The smartphone is worn on a belt at the lower back to record phone sensor data while walking. This data is used to identify foot strikes, calculate the total distance walked and step timing, and analyze pelvis accelerations. Information on symmetry, walking changes over time, and poor walking patterns is not available from a typical 2-6MWT and could help with clinical decision-making. The 2-6MWT app was evaluated in a pilot test using data from five able-bodied participants. Foot strike time was within 0.07 seconds when compared to gold standard video recordings. The total distance calculated by the app was within 1m of the measured distance.

Theory and analysis of frequency-domain photoacoustic tomography

A new frequency-domain approach to photoacoustic tomography has recently been proposed, promising to overcome some of the shortcomings associated with the pulsed photoacoustic approach. This approach offers many of the benefits of pulsed photoacoustics but requires a different set of equations for modeling of the forward and inverse problems due to the longer time scales involved in the optical input signal. The theory of photoacoustic tomography with an optical input that is not necessarily a short pulse is considered in this paper. The full optical, thermal, and acoustic governing equations are derived. A transfer function approach is taken for the solution and analysis of this problem. The results and implications are compared with those of pulsed photoacoustics and traditional ultrasonic diffraction tomography. A Fourier diffraction theorem is also presented, which could be used as a basis for the development of tomographic imaging algorithms.