@Article{Kulic2009_TRO,
  author    = {Kuli\'{c}, D. and Takano, W. and Nakamura, Y.},
  title     = {Online Segmentation and Clustering From Continuous Observation of Whole Body Motions},
  journal   = {IEEE Transactions on Robotics},
  year      = {2009},
  volume    = {25},
  pages     = {1158--1166},
  issn      = {1552-3098},
  abstract  = {This paper describes a novel approach for incremental learning of human motion pattern primitives through online observation of human motion. The observed time series data stream is first stochastically segmented into potential motion primitive segments, based on the assumption that data belonging to the same motion primitive will have the same underlying distribution. The motion segments are then abstracted into a stochastic model representation and automatically clustered and organized. As new motion patterns are observed, they are incrementally grouped together into a tree structure, based on their relative distance in the model space. The tree leaves, which represent the most specialized learned motion primitives, are then passed back to the segmentation algorithm so that as the number of known motion primitives increases, the accuracy of the segmentation can also be improved. The combined algorithm is tested on a sequence of continuous human motion data that are obtained through motion capture, and demonstrates the performance of the proposed approach.},
  doi       = {10.1109/TRO.2009.2026508},
  groups    = {STAT841, IROS2014, EMBC2014},
  keywords  = {clustering algorithm;human motion pattern primitives;humanoid robots;incremental learning;online observation;online segmentation;stochastic model representation;stochastic segmentation;time series data stream;tree structure;humanoid robots;image motion analysis;image segmentation;intelligent robots;learning (artificial intelligence);pattern clustering;robot vision;stochastic processes;time series;trees (mathematics);},
  review    = {- online, stocastic segmentation, unsupervisied, HMM, clustering, tree
- uses Kohlmorgen and Lemm's segmentation method
 - embed data stream to a higher dimension
 - take density distribution of a moving window
 - "temp state"
 - distance between windowed distribution is calculated
 - observation probability based on state distance is calculated with HMM
 - Viterbi used to determine state transision matrix
 - segment based on state transistion
- scaffolding
 - known motion is inserted as a "perm state"
 - if distance between input data and "perm state" are small, label it as 


Kuli\'{c} \etal \cite{Kulic2009a} extended Kohlmorgen's algorithm \cite{Kohlmorgen2002} by clustering together previously segmented sequences to generate new templates in real-time. Once a segment window has been identified, the segment is modeled as a HMM. The Kullback-Leibler distance between the observed HMM and existing models is calculated. If the distance is small, then the observation HMM is merged into the corresponding existing HMM. If not, it is added to the template collection, and used to improve the segmentation. The algorithm was verified on an 18-minute full-body motion sequence, and shows good segmentation accuracy, but also suffers from false positives due to the algorithm oversegmenting motion sequences into smaller subsequences that were considered to be a single segment by the manual segmentation.},
  timestamp = {2011.02.12},
}

{"_id":"mAaDQP7Nx6WSX6WeX","bibbaseid":"kuli-takano-nakamura-onlinesegmentationandclusteringfromcontinuousobservationofwholebodymotions-2009","downloads":0,"creationDate":"2017-09-14T16:34:36.685Z","title":"Online Segmentation and Clustering From Continuous Observation of Whole Body Motions","author_short":["Kulić, D.","Takano, W.","Nakamura, Y."],"year":2009,"bibtype":"article","biburl":"https://raw.githubusercontent.com/jfslin/jfslin.github.io/master/jf2lin.bib","bibdata":{"bibtype":"article","type":"article","author":[{"propositions":[],"lastnames":["Kulić"],"firstnames":["D."],"suffixes":[]},{"propositions":[],"lastnames":["Takano"],"firstnames":["W."],"suffixes":[]},{"propositions":[],"lastnames":["Nakamura"],"firstnames":["Y."],"suffixes":[]}],"title":"Online Segmentation and Clustering From Continuous Observation of Whole Body Motions","journal":"IEEE Transactions on Robotics","year":"2009","volume":"25","pages":"1158--1166","issn":"1552-3098","abstract":"This paper describes a novel approach for incremental learning of human motion pattern primitives through online observation of human motion. The observed time series data stream is first stochastically segmented into potential motion primitive segments, based on the assumption that data belonging to the same motion primitive will have the same underlying distribution. The motion segments are then abstracted into a stochastic model representation and automatically clustered and organized. As new motion patterns are observed, they are incrementally grouped together into a tree structure, based on their relative distance in the model space. The tree leaves, which represent the most specialized learned motion primitives, are then passed back to the segmentation algorithm so that as the number of known motion primitives increases, the accuracy of the segmentation can also be improved. The combined algorithm is tested on a sequence of continuous human motion data that are obtained through motion capture, and demonstrates the performance of the proposed approach.","doi":"10.1109/TRO.2009.2026508","groups":"STAT841, IROS2014, EMBC2014","keywords":"clustering algorithm;human motion pattern primitives;humanoid robots;incremental learning;online observation;online segmentation;stochastic model representation;stochastic segmentation;time series data stream;tree structure;humanoid robots;image motion analysis;image segmentation;intelligent robots;learning (artificial intelligence);pattern clustering;robot vision;stochastic processes;time series;trees (mathematics);","review":"- online, stocastic segmentation, unsupervisied, HMM, clustering, tree - uses Kohlmorgen and Lemm's segmentation method - embed data stream to a higher dimension - take density distribution of a moving window - \"temp state\" - distance between windowed distribution is calculated - observation probability based on state distance is calculated with HMM - Viterbi used to determine state transision matrix - segment based on state transistion - scaffolding - known motion is inserted as a \"perm state\" - if distance between input data and \"perm state\" are small, label it as Kulić \\etal i̧teKulic2009a extended Kohlmorgen's algorithm i̧teKohlmorgen2002 by clustering together previously segmented sequences to generate new templates in real-time. Once a segment window has been identified, the segment is modeled as a HMM. The Kullback-Leibler distance between the observed HMM and existing models is calculated. If the distance is small, then the observation HMM is merged into the corresponding existing HMM. If not, it is added to the template collection, and used to improve the segmentation. The algorithm was verified on an 18-minute full-body motion sequence, and shows good segmentation accuracy, but also suffers from false positives due to the algorithm oversegmenting motion sequences into smaller subsequences that were considered to be a single segment by the manual segmentation.","timestamp":"2011.02.12","bibtex":"@Article{Kulic2009_TRO,\n author = {Kuli\\'{c}, D. and Takano, W. and Nakamura, Y.},\n title = {Online Segmentation and Clustering From Continuous Observation of Whole Body Motions},\n journal = {IEEE Transactions on Robotics},\n year = {2009},\n volume = {25},\n pages = {1158--1166},\n issn = {1552-3098},\n abstract = {This paper describes a novel approach for incremental learning of human motion pattern primitives through online observation of human motion. The observed time series data stream is first stochastically segmented into potential motion primitive segments, based on the assumption that data belonging to the same motion primitive will have the same underlying distribution. The motion segments are then abstracted into a stochastic model representation and automatically clustered and organized. As new motion patterns are observed, they are incrementally grouped together into a tree structure, based on their relative distance in the model space. The tree leaves, which represent the most specialized learned motion primitives, are then passed back to the segmentation algorithm so that as the number of known motion primitives increases, the accuracy of the segmentation can also be improved. The combined algorithm is tested on a sequence of continuous human motion data that are obtained through motion capture, and demonstrates the performance of the proposed approach.},\n doi = {10.1109/TRO.2009.2026508},\n groups = {STAT841, IROS2014, EMBC2014},\n keywords = {clustering algorithm;human motion pattern primitives;humanoid robots;incremental learning;online observation;online segmentation;stochastic model representation;stochastic segmentation;time series data stream;tree structure;humanoid robots;image motion analysis;image segmentation;intelligent robots;learning (artificial intelligence);pattern clustering;robot vision;stochastic processes;time series;trees (mathematics);},\n review = {- online, stocastic segmentation, unsupervisied, HMM, clustering, tree\n- uses Kohlmorgen and Lemm's segmentation method\n - embed data stream to a higher dimension\n - take density distribution of a moving window\n - \"temp state\"\n - distance between windowed distribution is calculated\n - observation probability based on state distance is calculated with HMM\n - Viterbi used to determine state transision matrix\n - segment based on state transistion\n- scaffolding\n - known motion is inserted as a \"perm state\"\n - if distance between input data and \"perm state\" are small, label it as \n\n\nKuli\\'{c} \\etal \\cite{Kulic2009a} extended Kohlmorgen's algorithm \\cite{Kohlmorgen2002} by clustering together previously segmented sequences to generate new templates in real-time. Once a segment window has been identified, the segment is modeled as a HMM. The Kullback-Leibler distance between the observed HMM and existing models is calculated. If the distance is small, then the observation HMM is merged into the corresponding existing HMM. If not, it is added to the template collection, and used to improve the segmentation. The algorithm was verified on an 18-minute full-body motion sequence, and shows good segmentation accuracy, but also suffers from false positives due to the algorithm oversegmenting motion sequences into smaller subsequences that were considered to be a single segment by the manual segmentation.},\n timestamp = {2011.02.12},\n}\n\n","author_short":["Kulić, D.","Takano, W.","Nakamura, Y."],"key":"Kulic2009_TRO","id":"Kulic2009_TRO","bibbaseid":"kuli-takano-nakamura-onlinesegmentationandclusteringfromcontinuousobservationofwholebodymotions-2009","role":"author","urls":{},"keyword":["clustering algorithm;human motion pattern primitives;humanoid robots;incremental learning;online observation;online segmentation;stochastic model representation;stochastic segmentation;time series data stream;tree structure;humanoid robots;image motion analysis;image segmentation;intelligent robots;learning (artificial intelligence);pattern clustering;robot vision;stochastic processes;time series;trees (mathematics);"],"downloads":0},"search_terms":["online","segmentation","clustering","continuous","observation","whole","body","motions","kulić","takano","nakamura"],"keywords":["clustering algorithm;human motion pattern primitives;humanoid robots;incremental learning;online observation;online segmentation;stochastic model representation;stochastic segmentation;time series data stream;tree structure;humanoid robots;image motion analysis;image segmentation;intelligent robots;learning (artificial intelligence);pattern clustering;robot vision;stochastic processes;time series;trees (mathematics);"],"authorIDs":[],"dataSources":["iCsmKnycRmHPxmhBd"]}