@conference {Lin2016_HIPOCT, title = {Comparison of Kinematic and Dynamic Sensor Modalities and Derived Features for Human Motion Segmentation}, booktitle = {Proceedings of the IEEE/NIH Strategic Conference on Healthcare Innovations and Point of Care Technologies}, year = {2016}, pages = {109{\textendash}112}, author = {Lin, J. F. S. and Bonnet, V. and Joukov, V. and Venture, G. and Kuli{\'c}, D.} } @conference {Lin2016_ICHR, title = {Human Motion Segmentation using Cost Weights Recovered from Inverse Optimal Control}, booktitle = {Proceedings of the IEEE/RAS International Conference on Humanoid Robots}, year = {2016}, pages = {1107{\textendash}1113}, author = {Lin, J. F. S. and Bonnet, V. and Panchea, A. M. and Ramdani, N. and Venture, G. and Kuli{\'c}, D.} } @article {Lin2016_THMS, title = {Movement Primitive Segmentation for Human Motion Modeling: A Framework for Analysis}, journal = {IEEE Transactions on Human-Machine Systems}, volume = {46}, year = {2016}, pages = {325{\textendash}339}, abstract = {Movement primitive segmentation enables long sequences of human movement observation data to be segmented into smaller components, termed movement primitives, to facilitate movement identification, modeling, and learning. It has been applied to exercise monitoring, gesture recognition, human-machine interaction, and robot imitation learning. This paper proposes a segmentation framework to categorize and compare different segmentation algorithms considering segment definitions, data sources, application-specific requirements, algorithm mechanics, and validation techniques. The framework is applied to human motion segmentation methods by grouping them into online, semionline, and offline approaches. Among the online approaches, distance-based methods provide the best performance, while stochastic dynamic models work best in the semionline and offline settings. However, most algorithms to date are tested with small datasets, and algorithm generalization across participants and to movement changes remains largely untested.}, keywords = {Algorithm design and analysis, algorithm mechanics, application-specific requirements, Cameras, classification algorithms, Data collection, data sources, Databases, distance-based methods, exercise monitoring, gesture recognition, human motion modeling, human movement observation data sequences, human-machine interaction, image motion analysis, image segmentation, machine learning algorithms, Manuals, Motion segmentation, movement identification, movement learning, movement modeling, movement primitive segmentation, physiology, robot imitation learning, segment definitions, segmentation algorithms, stochastic dynamic models, time series analysis, validation techniques}, issn = {2168-2291}, doi = {10.1109/THMS.2015.2493536}, author = {Lin, J. F. S. and Karg, M. E. and Kuli{\'c}, D.} } @conference {Lin2016_SWIT, title = {Segmentation by Data Point Classification Applied to Forearm Surface EMG}, booktitle = {Proceedings of the EAI International Summit of Smart City 360}, volume = {166}, year = {2016}, pages = {153{\textendash}165}, publisher = {Springer International Publishing}, organization = {Springer International Publishing}, doi = {10.1007/978-3-319-33681-7_13}, author = {Lin, J. F. S. and Samadani, A. and Kuli{\'c}, D.} } @conference {Lin2014_ICHR, title = {Full-body Multi-primitive Segmentation using Classifiers}, booktitle = {Proceedings of the IEEE/RAS International Conference on Humanoid Robots}, year = {2014}, pages = {874{\textendash}880}, author = {Lin, J. F. S. and Joukov, V. and Kuli{\'c}, D.} } @conference {Lin2014_EMBC, title = {Human Motion Segmentation by Data Point Classification}, booktitle = {Proceedings of the International Conference of the IEEE Engineering in Medicine and Biology Society}, year = {2014}, pages = {9{\textendash}13}, author = {Lin, J. F. S. and Joukov, V. and Kuli{\'c}, D.} } @article {Lin2012_TNSRE, title = {On-line Segmentation of Human Motion for Automated Rehabilitation Exercise Analysis}, journal = {IEEE Transactions on Neural Systems and Rehabilitation Engineering}, volume = {22}, year = {2014}, pages = {168{\textendash}180}, abstract = {To enable automated analysis of rehabilitation movements, an approach for accurately identifying and segmenting movement repetitions is required. This paper proposes an approach for on-line, automated segmentation and identification of movement segments from continuous time-series data of human movement. The proposed approach uses a two stage identification and recognition process, based on velocity features and stochastic modeling of each motion to be identified. In the first stage, motion segment candidates are identified based on a characteristic sequence of velocity features such as velocity peaks and zero velocity crossings. In the second stage, Hidden Markov models are used to accurately identify segment locations from the identified candidates. The proposed approach is capable of on-line segmentation and identification, enabling interactive feedback in rehabilitation applications. The approach is validated on a 20 person rehabilitation movement dataset.}, doi = {10.1109/TNSRE.2013.2259640}, author = {Lin, J. F. S. and Kuli{\'c}, D.} } @conference {Lin2013_EMBC, title = {Human Pose Recovery for Rehabilitation Using Ambulatory Sensors}, booktitle = {Proceedings of the International Conference of the IEEE Engineering in Medicine and Biology Society}, year = {2013}, pages = {4799{\textendash}4802}, address = {Osaka, Japan}, author = {Lin, J. F. S. and Kuli{\'c}, D.} } @conference {Zhang2013, title = {IMU based Single Stride Identification of Humans}, booktitle = {Proceedings of the IEEE International Symposium on Robot and Human Interactive Communication}, year = {2013}, pages = {220-225}, author = {Zhang, T. and Karg, M. E. and Lin, J. F. S. and Kuli{\'c}, D. and Venture, G.} } @mastersthesis {Lin2012_MASc, title = {Automated Rehabilitation Exercise Motion Tracking}, year = {2012}, school = {University of À¶Ý®ÊÓƵ}, type = {masters}, abstract = {Current physiotherapy practice relies on visual observation of the patient for diagnosis and assessment. The assessment process can potentially be automated to improve accuracy and reliability. This thesis proposes a method to recover patient joint angles and automatically extract movement profiles utilizing small and lightweight body-worn sensors. Joint angles are estimated from sensor measurements via the extended Kalman filter (EKF). Constant-acceleration kinematics is employed as the state evolution model. The forward kinematics of the body is utilized as the measurement model. The state and measurement models are used to estimate the position, velocity and acceleration of each joint, updated based on the sensor inputs from inertial measurement units (IMUs). Additional joint limit constraints are imposed to reduce drift, and an automated approach is developed for estimating and adapting the process noise during on-line estimation. Once joint angles are determined, the exercise data is segmented to identify each of the repetitions. This process of identifying when a particular repetition begins and ends allows the physiotherapist to obtain useful metrics such as the number of repetitions performed, or the time required to complete each repetition. A feature-guided hidden Markov model (HMM) based algorithm is developed for performing the segmentation. In a sequence of unlabelled data, motion segment candidates are found by scanning the data for velocity-based features, such as velocity peaks and zero crossings, which match the pre-determined motion templates. These segment potentials are passed into the HMM for template matching. This two-tier approach combines the speed of a velocity feature based approach, which only requires the data to be differentiated, with the accuracy of the more computationally-heavy HMM, allowing for fast and accurate segmentation. The proposed algorithms were verified experimentally on a dataset consisting of 20 healthy subjects performing rehabilitation exercises. The movement data was collected by IMUs strapped onto the hip, thigh and calf. The joint angle estimation system achieves an overall average RMS error of 4.27 cm, when compared against motion capture data. The segmentation algorithm reports 78\% accuracy when the template training data comes from the same participant, and 74\% for a generic template.}, url = {hdl.handle.net/10012/7191}, author = {Lin, J. F. S.} } @article {Lin2012_PMEA, title = {Human Pose Recovery using Wireless Inertial Measurement Units}, journal = {Physiological Measurement}, volume = {33}, year = {2012}, pages = {2099{\textendash}2115}, doi = {10.1088/0967-3334/33/12/2099}, author = {Lin, J. F. S. and Kuli{\'c}, D.} } @conference {Lin2012_EMBC, title = {Segmenting Human Motion for Automated Rehabilitation Exercise Analysis}, booktitle = {Proceedings of the International Conference of the IEEE Engineering in Medicine and Biology Society}, year = {2012}, pages = {2881{\textendash}2884}, address = {San Diego, USA}, abstract = {This paper proposes an approach for the automated segmentation and identification of movement segments from continuous time series data of human movement, collected through motion capture of ambulatory sensors. The proposed approach uses a two stage identification and recognition process, based on velocity and stochastic modeling of each motion to be identified. In the first stage, motion segment candidates are identified based on a unique sequence of velocity features such as velocity peaks and zero velocity crossings. In the second stage, Hidden Markov models are used to accurately identify segment locations from the identified candidates. The approach is capable of on-line segmentation and identification, enabling interactive feedback in rehabilitation applications. The approach is validated on a rehabilitation movement dataset, and achieves a segmentation accuracy of 89\%.}, doi = {10.1109/EMBC.2012.6346565}, author = {Lin, J. F. S. and Kuli{\'c}, D.} } @conference {Lin2011_IROS, title = {Automatic Human Motion Segmentation and Identification using Feature Guided HMM for Physical Rehabilitation Exercises}, booktitle = {Proceedings of the IEEE International Conference on Intelligent Robots and Systems, Workshop on Robotics for Neurology and Rehabilitation}, year = {2011}, pages = {33{\textendash}36}, address = {San Francisco, CA}, abstract = {Fast and accurate motion segmentation and identification methods are required to enable real-time assessment and feedback for physical rehabilitation. Exercise motions exhibit cyclic patterns that can be characterized by simple features, such as zero-velocity crossings or velocity peaks. In this paper, these features are used as framing windows for simultaneous motion segmentation and identification via Hidden Markov models. Comparisons to other segmentation methods show that feature guiding increases the segmentation accuracy and greatly reduces the runtime needed to perform the segmentation.}, keywords = {motion primitive, Motion segmentation}, author = {Lin, J. F. S. and Kuli{\'c}, D.} } @conference {Lin2011_SAIS, title = {Motion Segmentation with Inertial Measurements using Feature Guided HMM}, booktitle = {Proceedings of the Symposium on Advanced Intelligent Systems}, year = {2011}, abstract = {Motion segmentation and identification methods tend to assume that joint angle data are readily available for processing. The standard data acquisition method, motion capture, is not feasible for real-world deployment, such as in physical rehabilitation clinics, due to space, environment and cost constraints. In this paper, a feature-guided Hidden Markov model segmentation algorithm is adapted to segment and identify human motion from acceleration and gyroscopic data instead of position data. Acceleration and gyroscopic data are readily obtainable through small lightweight sensors, and would serve as a more appropriate sensor system for a rehabilitation clinic than a motion capture system.}, author = {Lin, J. F. S. and Kuli{\'c}, D.} }