Matthew Field

Recognizing human motions through mixture modeling of inertial data

Systems that recognize patterns in human motion are central to improvements in automation and human computer interaction. This work addresses challenges which arise in the context of recognizing arbitrary human actions from body-worn sensors. Chiefly the invariance to temporal scaling of events, coping with unlabeled data and estimating an appropriate model complexity. In order to deal with the severe case of unlabeled data, a method is proposed based on dynamic time alignment of Gaussian mixture model clusters for matching actions in an unsupervised temporal segmentation. In facilitation of this, an extensive corpus of continuous motion sequences composed of everyday tasks was recorded as analysis scenarios. The technique achieved an average accuracy of 72% for correctly merging actions performed by different participants. With labeled data and recognition models designed for particular classes, an accuracy of 89% was achieved in classifying the motion of participants left out of the modeling process. These results are contrasted with benchmark methods for recognition in a systematic validation revealing, in particular, an improved performance for mixture model prediction utilizing segments. HighlightsA method is proposed for unsupervised segment clustering of human motion capture data.Gaussian mixture models and dynamic time warping are used to compare similar data sequences.Human motion capture data was collected with a set of body-worn inertial sensors.The resultant classifier is compared with k-nearest-neighbor and support vector machine approaches.

Human motion capture sensors and analysis in robotics

Purpose – The purpose of this paper is to provide a review of various motion capture technologies and discuss the methods for handling the captured data in applications related to robotics.Design/methodology/approach – The approach taken in the paper is to compare the features and limitations of motion trackers in common use. After introducing the technology, a summary is given of robotic‐related work undertaken with the sensors and the strengths of different approaches in handling the data are discussed. Each comparison is presented in a table. Results from the authors experimentation with an inertial motion capture system are discussed based on clustering and segmentation techniques.Findings – The trend in methodology is towards stochastic machine learning techniques such as hidden Markov model or Gaussian mixture model, their extensions in hierarchical forms and non‐linear dimension reduction. The resulting empirical models tend to handle uncertainty well and are suitable for incrementally updating mo...

Motion capture in robotics review

This survey reviews motion capture technologies and the current challenges associated with their application in robotic systems. Various sensor systems used in current literature are introduced and evaluated based on the relative strengths and weaknesses. Some research problems pursued with these sensors in robotics are reviewed and application areas are discussed. Significant methodologies in analysing the sensor data are discussed and evaluated based on the perceived benefits and limitations. Finally, results from experimentation with an inertial motion capture system are shown based on clustering and segmentation techniques.

Learning Trajectories for Robot Programing by Demonstration Using a Coordinated Mixture of Factor Analyzers

This paper presents an approach for learning robust models of humanoid robot trajectories from demonstration. In this formulation, a model of the joint space trajectory is represented as a sequence of motion primitives where a nonlinear dynamical system is learned by constructing a hidden Markov model (HMM) predicting the probability of residing in each motion primitive. With a coordinated mixture of factor analyzers as the emission probability density of the HMM, we are able to synthesize motion from a dynamic system acting along a manifold shared by both demonstrator and robot. This provides significant advantages in model complexity for kinematically redundant robots and can reduce the number of corresponding observations required for further learning. A stability analysis shows that the system is robust to deviations from the expected trajectory as well as transitional motion between manifolds. This approach is demonstrated experimentally by recording human motion with inertial sensors, learning a motion primitive model and correspondence map between the human and robot, and synthesizing motion from the manifold to control a 19 degree-of-freedom humanoid robot.

Objective Functional Capacity Assessment Using Inertial Sensor

Functional capacity assessment is carried out to measure the functional limitations of a subject. While the clinical assessment can be validated against various standards, quantifying the assessment and achieving an objective, repeatable, and reliable score in the clinical assessment is a challenge. Current methods are subjective. The Progressive Iso inertial Lifting Evaluation (PILE) is a lifting test developed for functional capacity assessment. The primary aim of this study is to improve reliability and repeatability of PILE through objective measurement of patients performance. This is achieved by recording and analysing the movement of a patient by a motion capture system based on a network array of inertial wireless sensors. Various analyses conducted on the data indicate that the captured data provides adequate information to objectively determine the failure of the subject to maintain correct posture and to identify the onset of muscle fatigue within the PILE assessment.

Mixture Model Segmentation for Gait Recognition

Modeling of human motion through a discrete sequence of motion primitives, retaining elements of skillful or unique motion of an individual is addressed. Using wireless inertial motion sensors, a skeletal model of the fluid human gait was gathered. The posture of the human model is described by sets of Euler angles for each sample. An intrinsic classification algorithm known as minimum message length encoding (MML) is deployed to segment the stream of data and subsequently formulate certain Gaussian mixture models (GMM) that contain a plausible range of motion primitives. The removal of certain less seemingly important modes has been shown to significantly affect the fluidity of a gait cycle. The approach is described and the outcomes so far are provided.

Inertial sensing for human motor control symmetry in injury rehabilitation

This paper proposes a series of methods for representing changes in human motion during injury rehabilitation using Micro-Electro-Mechanical Systems (MEMS) inertial sensors. Tracking the changes over a recovery period requires methods for evaluating the similarity of movement in an impaired state against a non-impaired state. We investigate the use of motion analyses such as the centre of mass (COM) tipping distance, the variance of joint velocity eigenvalues and the cumulative state changes of Gaussian mixture models (GMM) for monitoring the symmetry between the left and right sides of body during rehabilitation exercises. The methods are tested on an injured athlete over 4 months of recovery from an ankle operation and validated by comparing the observed improvement to the variation among a group of uninjured subjects. The results indicate that gradual changes are detected in the motion symmetry, thus providing quantitative measures to aid clinical decisions.

A prediction model for early death in non-small cell lung cancer patients following curative-intent chemoradiotherapy

Abstract Background: Early death after a treatment can be seen as a therapeutic failure. Accurate prediction of patients at risk for early mortality is crucial to avoid unnecessary harm and reducing costs. The goal of our work is two-fold: first, to evaluate the performance of a previously published model for early death in our cohorts. Second, to develop a prognostic model for early death prediction following radiotherapy. Material and methods: Patients with NSCLC treated with chemoradiotherapy or radiotherapy alone were included in this study. Four different cohorts from different countries were available for this work (N = 1540). The previous model used age, gender, performance status, tumor stage, income deprivation, no previous treatment given (yes/no) and body mass index to make predictions. A random forest model was developed by learning on the Maastro cohort (N = 698). The new model used performance status, age, gender, T and N stage, total tumor volume (cc), total tumor dose (Gy) and chemotherapy timing (none, sequential, concurrent) to make predictions. Death within 4 months of receiving the first radiotherapy fraction was used as the outcome. Results: Early death rates ranged from 6 to 11% within the four cohorts. The previous model performed with AUC values ranging from 0.54 to 0.64 on the validation cohorts. Our newly developed model had improved AUC values ranging from 0.62 to 0.71 on the validation cohorts. Conclusions: Using advanced machine learning methods and informative variables, prognostic models for early mortality can be developed. Development of accurate prognostic tools for early mortality is important to inform patients about treatment options and optimize care.

Nonlinear bilateral teleoperation using extended active observer for force estimation and disturbance suppression

A novel nonlinear teleoperation algorithm for simultaneous inertial parameters and force estimation at the master and slave sides of the teleoperation system is proposed. The scheme, called Extended Active Observer (EAOB), is an extension of the existing active observer. It provides effective force tracking at the master side with accurate position tracking at the slave side in the presence of inertial parameter variation and measurement noise. The proposed method only requires the measurement of robot position, and as a result significantly reduces the difficulty and cost of implementing bilateral teleoperation systems. The approach is described and its stability is analytically verified. The performance of the method is validated through computer simulation and compared with the Nicosia observer-based controller. According to the results, EAOB outperforms the Nicosia observer method and effectively rejects noise.

Modelling dynamic body motion using IMUs positioned on the torso

This study investigated the application of Symbolic Aggregate approXimation (SAX) to modelling dynamic body motion using a single inertial measurement unit (IMU). In addition this study demonstrates how IMUs located at different positions around the body produce comparable results. This study investigates the output of multiple IMU sensors, employed to monitor movement. Next a comparison of the sternum, pelvis, head and lower back sensor locations is conducted by analysing the measured rotation and position IMU data. Additionally, the classifier has been improved by increasing the information in the training data to avoid incorrect classification of similar activities. The results obtained in this study also prove that the sternum and head sensors provided comparable data to the pelvis sensor when using TSBs for classification, especially when used to classify dynamic activities. To pre-process the data, sub-dimensional motif discovery was employed to find features within the data from multiple IMUs. This improves on previous studies which illustrated difficulty classifying fast movements using the sternum IMU. This data was also approximated using SAX and classified by comparing Time Series Bitmaps (TSBs) to find the least Euclidean distance between the reference TSBs and the sliding window TSBs.