James Y. Xu

Integrated Inertial Sensors and Mobile Computing for Real-Time Cycling Performance Guidance via Pedaling Profile Classification

Today, the bicycle is utilized as a daily commute tool, a physical rehabilitation asset, and sporting equipment, prompting studies into the biomechanics of cycling. Of the number of important parameters that affect cycling efficiency, the foot angle profile is one of the most important as it correlates directly with the effective force applied to the bike. However, there has been no compact and portable solution for measuring the foot angle and for providing the cyclist with real-time feedback due to a number of difficulties of the current tracking and sensing technologies and the myriad types of bikes available. This paper presents a novel sensing and mobile computing system for classifying the foot angle profiles during cycling and for providing real-time guidance to the user to achieve the correct profile. Continuous foot angle tracking is firstly converted into a discrete problem requiring only recognition of acceleration profiles of the foot using a single shoe mounted tri-axial accelerometer during each pedaling cycle. A classification method is then applied to identify the pedaling profile. Finally, a mobile solution is presented to provide real-time signal processing and guidance.

Context-driven, Prescription-Based Personal Activity Classification: Methodology, Architecture, and End-to-End Implementation

Enabling large-scale monitoring and classification of a range of motion activities is of primary importance due to the need by healthcare and fitness professionals to monitor exercises for quality and compliance. Past work has not fully addressed the unique challenges that arise from scaling. This paper presents a novel end-to-end system solution to some of these challenges. The system is built on the prescription-based context-driven activity classification methodology. First, we show that by refining the definition of context, and introducing the concept of scenarios, a prescription model can provide personalized activity monitoring. Second, through a flexible architecture constructed from interface models, we demonstrate the concept of a context-driven classifier. Context classification is achieved through a classification committee approach, and activity classification follows by means of context specific activity models. Then, the architecture is implemented in an end-to-end system featuring an Android application running on a mobile device, and a number of classifiers as core classification components. Finally, we use a series of experimental field evaluations to confirm the expected benefits of the proposed system in terms of classification accuracy, rate, and sensor operating life.

Gait analysis using 3D motion reconstruction with an activity-specific tracking protocol

In this paper, we present a new gait analysis method using 3D body motion reconstruction with an activity-specific tracking protocol. A kinematic chain modeling the movement of lower extremities was constructed for general lower body activity monitoring. By exploring the nature of walking, a constrained forward-backward statistical linearized sigma-point Kalman Smoother with periodic state vector resetting was developed. This tracks the dynamic joint configuration during walking. Direct experimental evaluation was provided by step length computation as well as complete motion reconstruction. This method has demonstrated stable long term tracking of walking and yields greater than 95% accuracy for step length estimation.

Context guided and personalized activity classification system

Continued rapid progress in the development of embedded motion sensing enables wearable devices that provide fundamental advances in the capability to monitor and classify human motion, detect movement disorders, and estimate energy expenditure. With this progress, it is becoming possible to provide, for the first time, evaluation of outcomes of rehabilitation interventions and direct guidance for advancement of subject health, wellness, and safety. The progress in motion classification relies on both the performance of new sensor fusion methods that provide inference, and the energy efficiency of energy-constrained monitoring sensors. As will be described here, both of these objectives require advances in the capability of detecting and classifying the location and environmental context. Context directly enables both enhanced motion classification accuracy and speed through reduction in search space, and reduced energy demand through context-aware optimization of sensor sampling and operation schedules. There have been attempts to introduce context awareness into activity monitoring with limited success, due to the ambiguity in the definition of context, and the lack of a system architecture that enables the adaptation of signal processing and sensor fusion algorithms specific to the task of personalized activity monitoring. In this paper we present a novel end-to-end system that provides context guided personalized activity classification. With a refined concept of context, the system introduces interface models that feature a context classification committee, the concept of context specific activity classification, the ability to manage sensors given context, and the ability to operate in real time through web services. We also present an implementation that demonstrates accurate context classification, accurate activity classification using context specific models with improved accuracy and speed, and extended operating life through sensor energy management.

Personalized Active Learning for Activity Classification Using Wireless Wearable Sensors

Enabling accurate and low-cost classification of a range of motion activities is important for numerous applications, ranging from disease treatment and in-community rehabilitation of patients to athlete training. This paper proposes a novel contextual online learning method for activity classification based on data captured by low-cost, body-worn inertial sensors, and smartphones. The proposed method is able to address the unique challenges arising in enabling online, personalized and adaptive activity classification without requiring training phase from the individual. Another key challenge of activity classification is that the labels may change over time, as the data as well as the activity to be monitored evolve continuously, and the true label is often costly and difficult to obtain. The proposed algorithm is able to actively learn when to ask for the true label by assessing the benefits and costs of obtaining them. We rigorously characterize the performance of the proposed learning algorithm and Our experiments show that the proposed algorithm outperforms existing algorithms.

Enabling Large-Scale Ground-Truth Acquisition and System Evaluation in Wireless Health

Large-scale activity monitoring is a core component of systems aiming to improve our ability to manage fitness, deliver care, and diagnose conditions. While much research has been devoted to the accurate classification of motion, the challenges arising from scaling to large communities have received little attention. This paper introduces the problem of scaling, and addresses two of the most important issues: enabling robust large-scale ground-truth acquisition and building a common database for systems comparison. This paper presents a voice powered mobile acquisition system with efficient annotation tools and an extendable online searchable activity database with 331 datasets totaling over 700 h with 8 sensing modalities and 15 activities.

Context-guided universal hybrid decision tree for activity classification

Obtaining accurate measurements of human activities is important for a broad set of health applications. We propose a context-based hybrid decision tree classifier with a real-time portable solution for reliably classifying daily life activities and for providing instant feedback. At first, to determine user contexts, we utilize sensors typically found on smart phones or tablets to collect environment data. Then, we select different types of hybrid decision tree classifiers based on detected human context. The tree classifier can flexibly implement different decision rules at its internal nodes, and can be adapted from a population-based model when supplemented by training data for individuals. In addition, with the introduction of portable devices, the users can receive instant feedback of their current mobility status.

Web-Based Billing System Exploits Mature and Emerging Technology

Combining REST and Ajax with the traditional model-view-controller software architecture and using emerging technologies in software development, such as object relational modeling and Python, results in a successful new Web-based billing system.

Context-driven online learning for activity classification in wireless health

Enabling accurate and low-cost classification of a range of motion activities is of significant importance for wireless health through body worn inertial sensors and smartphones, due to the need by healthcare and fitness professonals to monitor exercises for quality and compliance. This paper proposes a novel contextual multi-armed bandits approach for large-scale activity classification. The proposed method is able to address the unique challenges arising from scaling, lack of training data and adaptation by melding context augmentation and continuous online learning into traditional activity classification. We rigorously characterize the performance of the proposed learning algorithm and prove that the learning regret (i.e. reward loss) is sublinear in time, thereby ensuring fast convergence to the optimal reward as well as providing short-term performance guarantees. Our experiments show that the proposed algorithm outperforms existing algorithms in terms of both providing higher classification accuracy as well as lower energy consumption.