David Howard

A Comparison of Feature Extraction Methods for the Classification of Dynamic Activities From Accelerometer Data

Driven by the demands on healthcare resulting from the shift toward more sedentary lifestyles, considerable effort has been devoted to the monitoring and classification of human activity. In previous studies, various classification schemes and feature extraction methods have been used to identify different activities from a range of different datasets. In this paper, we present a comparison of 14 methods to extract classification features from accelerometer signals. These are based on the wavelet transform and other well-known time- and frequency-domain signal characteristics. To allow an objective comparison between the different features, we used two datasets of activities collected from 20 subjects. The first set comprised three commonly used activities, namely, level walking, stair ascent, and stair descent, and the second a total of eight activities. Furthermore, we compared the classification accuracy for each feature set across different combinations of three different accelerometer placements. The classification analysis has been performed with robust subject-based cross-validation methods using a nearest-neighbor classifier. The findings show that, although the wavelet transform approach can be used to characterize nonstationary signals, it does not perform as accurately as frequency-based features when classifying dynamic activities performed by healthy subjects. Overall, the best feature sets achieved over 95% intersubject classification accuracy.

Inertial sensor-based knee flexion/extension angle estimation

A new method for estimating knee joint flexion/extension angles from segment acceleration and angular velocity data is described. The approach uses a combination of Kalman filters and biomechanical constraints based on anatomical knowledge. In contrast to many recently published methods, the proposed approach does not make use of the earths magnetic field and hence is insensitive to the complex field distortions commonly found in modern buildings. The method was validated experimentally by calculating knee angle from measurements taken from two IMUs placed on adjacent body segments. In contrast to many previous studies which have validated their approach during relatively slow activities or over short durations, the performance of the algorithm was evaluated during both walking and running over 5 minute periods. Seven healthy subjects were tested at various speeds from 1 to 5 mile/h. Errors were estimated by comparing the results against data obtained simultaneously from a 10 camera motion tracking system (Qualysis). The average measurement error ranged from 0.7 degrees for slow walking (1 mph) to 3.4 degrees for running (5 mph). The joint constraint used in the IMU analysis was derived from the Qualysis data. Limitations of the method, its clinical application and its possible extension are discussed.

An In Vivo Experimental Validation of a Computational Model of Human Foot

Reliable computational foot models offer an alternative means to enhance knowledge on the biomechanics of human foot. Model validation is one of the most critical aspects of the entire foot modeling and analysis process. This paper presents an in vivo experiment combining motion capture system and plantar pressure measure platform to validate a three-dimensional finite element model of human foot. The Magnetic Resonance Imaging (MRI) slices for the foot modeling and the experimental data for validation were both collected from the same volunteer subject. The validated components included the comparison of static model predictions of plantar force, plantar pressure and foot surface deformation during six loading conditions, to equivalent measured data. During the whole experiment, foot surface deformation, plantar force and plantar pressure were recorded simultaneously during six different loaded standing conditions. The predictions of the current FE model were in good agreement with these experimental results.

Dimensional change in muscle as a control signal for powered upper limb prostheses: a pilot study

The vast majority of externally powered prostheses are controlled from the myoelectric signal, measured at the skin surface using socket-located electrodes. This signal has been well researched and sophisticated signal processing methods developed. Nevertheless, the inherent properties of the signal, such as its broad bandwidth and low voltage amplitude, make its use less than straightforward in the control of low frequency activity such as powered prosthetic hand movement. This paper reports on a pilot study of an alternative, a signal derived from dimensional change in muscle. A new socket-located sensor was designed to measure dimensional change in muscle, the linearised output of which is termed the myokinemetric (MK) signal. This was used in a series of tasks aimed at investigating the potential for its use in upper-limb prosthesis control. Six amputee subjects were tested, of whom one was a regular user of the myoelectric hand, one had some experience, and four had little or no previous experience of controlling devices using their residual limb. Data is presented on the problems of shift in signal range with time and socket donning and doffing and on the ability of subjects to control the amplitude of the signal. The results show that subjects were able to control the magnitude of the MK signal to a significant degree, with typical errors averaging 0.1-0.3 mm, around 10% of the signal range. The principal problem encountered was the shift in signal with time and socket donning and doffing.

Error in the description of foot kinematics due to violation of rigid body assumptions

Kinematic data from rigid segment foot models inevitably includes errors because the bones within each segment move relative to each other. This study sought to define error in foot kinematic data due to violation of the rigid segment assumption. The research compared kinematic data from 17 different mid and forefoot rigid segment models to kinematic data of the individual bones comprising these segments. Kinematic data from a previous dynamic cadaver model study was used to derive individual bone as well as foot segment kinematics. Mean and maximum errors due to violation of the rigid body assumption varied greatly between models. The model with least error was the combination of navicular and cuboid (mean errors < = 1.3 degrees, average maximum error < = 2.4 degrees). Greatest error was seen for the model combining all the ten bones (mean errors < = 4.4 degrees, average maximum errors < = 6.9 degrees). Based on the errors reported a three segment mid and forefoot model is proposed: (1) Navicular and cuboid, (2) cuneiforms and metatarsals 1, 2 and 3, and (3) metatarsals 4 and 5. However the utility of this model will depend on the precise purpose of the in vivo foot kinematics research study being undertaken.

Automated Detection of Instantaneous Gait Events Using Time Frequency Analysis and Manifold Embedding

Accelerometry is a widely used sensing modality in human biomechanics due to its portability, non-invasiveness, and accuracy. However, difficulties lie in signal variability and interpretation in relation to biomechanical events. In walking, heel strike and toe off are primary gait events where robust and accurate detection is essential for gait-related applications. This paper describes a novel and generic event detection algorithm applicable to signals from tri-axial accelerometers placed on the foot, ankle, shank or waist. Data from healthy subjects undergoing multiple walking trials on flat and inclined, as well as smooth and tactile paving surfaces is acquired for experimentation. The benchmark timings at which heel strike and toe off occur, are determined using kinematic data recorded from a motion capture system. The algorithm extracts features from each of the acceleration signals using a continuous wavelet transform over a wide range of scales. A locality preserving embedding method is then applied to reduce the high dimensionality caused by the multiple scales while preserving salient features for classification. A simple Gaussian mixture model is then trained to classify each of the time samples into heel strike, toe off or no event categories. Results show good detection and temporal accuracies for different sensor locations and different walking terrains.

Regression techniques for the prediction of lower limb kinematics.

This work presents a novel and extensive investigation of mathematical regression techniques, for the prediction of laboratory-type kinematic measurements during human gait, from wearable measurement devices, such as gyroscopes and accelerometers. Specifically, we examine the hypothesis of predicting the segmental angles of the legs (left and right foot, shank and thighs), from rotational foot velocities and translational foot accelerations. This first investigation is based on kinematic data emulated from motion-capture laboratory equipment. We employ eight established regression algorithms with different properties, ranging from linear methods and neural networks with polynomial support and expanded nonlinearities, to radial basis functions, nearest neighbors and kernel density methods. Data from five gait cycles of eight subjects are used to perform both inter-subject and intra-subject assessments of the prediction capabilities of each algorithm, using cross-validation resampling methods. Regarding the algorithmic suitability to gait prediction, results strongly indicate that nonparametric methods, such as nearest neighbors and kernel density based, are particularly advantageous. Numerical results show high average prediction accuracy (rho = 0.98/0.99, RMS = 5.63 degrees/2.30 degrees, MAD = 4.43 degrees/1.52 degrees for inter/intra-subject testing). The presented work provides a promising and motivating investigation on the feasibility of cost-effective wearable devices used to acquire large volumes of data that are currently collected only from complex laboratory environments.

Automated design of robust discriminant analysis classifier for foot pressure lesions using kinematic data

In the recent years, the use of motion tracking systems for acquisition of functional biomechanical gait data, has received increasing interest due to the richness and accuracy of the measured kinematic information. However, costs frequently restrict the number of subjects employed, and this makes the dimensionality of the collected data far higher than the available samples. This paper applies discriminant analysis algorithms to the classification of patients with different types of foot lesions, in order to establish an association between foot motion and lesion formation. With primary attention to small sample size situations, we compare different types of Bayesian classifiers and evaluate their performance with various dimensionality reduction techniques for feature extraction, as well as search methods for selection of raw kinematic variables. Finally, we propose a novel integrated method which fine-tunes the classifier parameters and selects the most relevant kinematic variables simultaneously. Performance comparisons are using robust resampling techniques such as Bootstrap 632+ and k-fold cross-validation. Results from experimentations with lesion subjects suffering from pathological plantar hyperkeratosis, show that the proposed method can lead to /spl sim/96% correct classification rates with less than 10% of the original features.

A generic analytical foot rollover model for predicting translational ankle kinematics in gait simulation studies

The objective of this paper is to develop an analytical framework to representing the ankle-foot kinematics by modelling the foot as a rollover rocker, which cannot only be used as a generic tool for general gait simulation but also allows for case-specific modelling if required. Previously, the rollover models used in gait simulation have often been based on specific functions that have usually been of a simple form. In contrast, the analytical model described here is in a general form that the effective foot rollover shape can be represented by any polar function rho=rho(phi). Furthermore, a normalized generic foot rollover model has been established based on a normative foot rollover shape dataset of 12 normal healthy subjects. To evaluate model accuracy, the predicted ankle motions and the centre of pressure (CoP) were compared with measurement data for both subject-specific and general cases. The results demonstrated that the ankle joint motions in both vertical and horizontal directions (relative RMSE approximately 10%) and CoP (relative RMSE approximately 15% for most of the subjects) are accurately predicted over most of the stance phase (from 10% to 90% of stance). However, we found that the foot cannot be very accurately represented by a rollover model just after heel strike (HS) and just before toe off (TO), probably due to shear deformation of foot plantar tissues (ankle motion can occur without any foot rotation). The proposed foot rollover model can be used in both inverse and forward dynamics gait simulation studies and may also find applications in rehabilitation engineering.

A Phase-Dependent Hypothesis for Locomotor Functions of Human Foot Complex

The human foot is a very complex structure comprising numerous bones, muscles, ligaments and synovial joints. As the only component in contact with the ground, the foot complex delivers a variety of biomechanical functions during human locomotion, e.g. body support and propulsion, stability maintenance and impact absorption. These need the human foot to be rigid and damped to transmit ground reaction forces to the upper body and maintain body stability, and also to be compliant and resilient to moderate risky impacts and save energy. How does the human foot achieve these apparent conflicting functions? In this study, we propose a phase-dependent hypothesis for the overall locomotor functions of the human foot complex based on in-vivo measurements of human natural gait and simulation results of a mathematical foot model. We propse that foot functions are highly dependent on gait phase, which is a major characteristics of human locomotion. In early stance just after heel strike, the foot mainly works as a shock absorber by moderating high impacts using the viscouselastic heel pad in both vertical and horizontal directions. In mid-stance phase (~80% of stance phase), the foot complex can be considered as a springy rocker, reserving external mechanical work using the foot arch whilst moving ground contact point forward along a curved path to maintain body stability. In late stance after heel off, the foot complex mainly serves as a force modulator like a gear box, modulating effective mechanical advantages of ankle plantiflexor muscles using metatarsal-phalangeal joints. A sound understanding of how diverse functions are implemented in a simple foot segment during human locomotion might be useful to gain insight into the overall foot locomotor functions and hence to facilitate clinical diagnosis, rehabilitation product design and humanoid robot development.