CoChat
Sign Up
Cited by 131Context-Aware Activity Recognition SystemsHuman Pose and Action RecognitionHuman Mobility and Location-Based Analysis

Activity recognition using semi-Markov models on real world smart home datasets

Open access
van Kasteren, T.L.M., Englebienne, G., Kröse, B.J.A.·2010·Journal of Ambient Intelligence and Smart Environments
DOI 10.3233/ais-2010-0070
Open this paper in CoChat — free accountOpen PDF

Abstract

Accurately recognizing human activities from sensor data recorded in a smart home setting is a challenging task. Typically, probabilistic models such as the hidden Markov model (HMM) or conditional random fields (CRF) are used to map the observed sen

Integrity signals

Registry

Crossref · 10.3233/ais-2010-0070

Cites

30 of 30 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

6 of 24 cited sources checked · 4 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    A lot of work is evaluated on laboratory datasets and simulated data.

    “A lot of work is evaluated on laboratory datasets [5,14,24] and simulated data [9,11].”
    Quote in the paperSubstantial claim1 of 5 sources checked
    • cites [5] 5. T. Duong, D. Phung, H. Bui, and S. Venkatesh. Efficient dura- tion and hierarchical modeling for human activity recognition. Artif. Intell., 173(7-8):830-856, 2009. source
    • cites [14] 14. D.L. Vail. Conditional Random Fields for Activity Recogni- tion. PhD Thesis, Carnegie Mellon University, 2008.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [24] 24. T.T. Truyen, H.H. Bui, and S. Venkatesh. Human activity learning and segmentation using partially hidden discrimina- tive models. In HAREM, 2005.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [9] 9. B. Logan, J. Healey, M. Philipose, E. Munguia-Tapia, and S. Intille. A long-term evaluation of sensing modalities for ac- tivity recognition. In Ubicomp, 2007.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [11] 11. K.P. Murphy. Hidden semi-Markov models (hsmms). Techni- cal report, University of British Columbia, 2002.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Findingown data

    In the paper's experiments, semi-Markov conditional random fields (SMCRFs) only slightly outperform conditional random fields (CRFs).

    “SMCRFs only slightly outperform CRFs, showing that CRFs are more robust in dealing with violations of the modelling assumptions.”
    Quote in the paperSubstantial claim
  • Findingown data

    In the paper's experiments, the hidden semi-Markov model (HSMM) consistently outperforms the hidden Markov model (HMM) on real-world smart home activity recognition datasets.

    “In our experiments the HSMM consistently outperforms the HMM, showing that accurate duration modelling can result in a significant increase in recognition performance.”
    Quote in the paperSubstantial claim
  • Findingown data

    SMCRFs outperform HMMs in terms of precision and recall on all four real world datasets.

    “They show that HSMMs, CRFs and SMCRFs outperform HMMs in terms of precision and recall on all datasets.”
    Quote in the paperSubstantial claim
  • Findingown data

    CRFs significantly outperform HMMs in terms of recall and F-measure on the kitchen2 dataset.

    “CRFs and SMCRFs significantly outperform HMMs in all measures on the 'bathroom1' dataset and in terms of recall and F-measure on the 'kitchen2' dataset.”
    Quote in the paperSubstantial claim
  • Classified as probably not a substantial claim. 12
  • May combine two claims. 13

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Hierarchical models can model a state using a number of substates in which the duration of the state is a convolution of geometric distributions.

    “Hierarchical models can model a state using a number of substates in which the duration of the state is a convolution of geometric distributions [18].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [18] 18. F. Sha and F. Pereira. Shallow parsing with conditional random fields. In NAACL '03: Proceedings of the 2003 Conference of the North American Chapter of the Association for Com- putational Linguistics on Human Language Technology, pages 134-141, Morristown, NJ, USA, 2003. Association for Com- putational Linguistics. source

      The Introduction mentions multilevel Markov models, and §4.2 describes paired chunk tags in a CRF. Neither passage addresses substates or a state-duration distribution; searches for the claimed convolution of geometric distributions and for hierarchical models with substates found no supporting passage in our copy. Page numbers were not available. (LLM, conf=0.73)

      Experimental — accuracy not yet measured

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    The Coxian distribution can perform equally well as other distributions, but can save significant computational cost.

    “Previous work has shown that the Coxian distribution can perform equally well as other distributions, but can save significant computational cost [5].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [5] 5. T. Duong, D. Phung, H. Bui, and S. Venkatesh. Efficient dura- tion and hierarchical modeling for human activity recognition. Artif. Intell., 173(7-8):830-856, 2009. source

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Duong et al. applied HSMMs to activity recognition and compared performance using various duration distributions, suggesting the Coxian distribution for computational efficiency.

    “In work by Duong et al.HSMMs were applied to activity recognition [5].They compared performance of the model using var-ious duration distributions and suggested the use of the Coxian distribution for computational efficiency.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [5] 5. T. Duong, D. Phung, H. Bui, and S. Venkatesh. Efficient dura- tion and hierarchical modeling for human activity recognition. Artif. Intell., 173(7-8):830-856, 2009. source

  • May combine two claims.

    Premisecitation

    Previous work has separately evaluated HSMMs and SMCRFs on simulated and laboratory data, but this work is the first to evaluate these models on real-world activity recognition data and compare their performance in a single paper.

    “Previous work has separately evaluated HSMMs [5] and SMCRFs [11,24] on simulated and laboratory data, but our work is the first to evaluate these models on real world activity recognition data and compare their performance in a single paper.”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites [5] 5. T. Duong, D. Phung, H. Bui, and S. Venkatesh. Efficient dura- tion and hierarchical modeling for human activity recognition. Artif. Intell., 173(7-8):830-856, 2009. source
    • cites [11] 11. K.P. Murphy. Hidden semi-Markov models (hsmms). Techni- cal report, University of British Columbia, 2002.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [24] 24. T.T. Truyen, H.H. Bui, and S. Venkatesh. Human activity learning and segmentation using partially hidden discrimina- tive models. In HAREM, 2005.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Semi-Markov models can significantly outperform hierarchical models, despite their better duration modelling.

    “It has been shown that semi-Markov models can significantly outperform hierarchical models, despite their better duration modelling [5].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [5] 5. T. Duong, D. Phung, H. Bui, and S. Venkatesh. Efficient dura- tion and hierarchical modeling for human activity recognition. Artif. Intell., 173(7-8):830-856, 2009. source

      Section 5.4.3 (page unavailable) reports that the Coxian switching hidden semi-Markov model significantly outperformed a standard hierarchical HMM in activity segmentation. But the winning model combines semi-Markov duration modeling with a hierarchical structure; the result does not establish that a non-hierarchical semi-Markov model outperforms hierarchical models generally. Section 2.2 also says hierarchical HMMs lack explicit duration models, rather than having better duration modeling. (LLM, conf=0.83)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingown data

    At a 95% confidence interval, HSMM significantly outperforms HMM on all datasets and in all measures.

    “At a confidence interval of 95% the HSMM significantly outperforms the HMM on all the datasets and in all the measures.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Findingnone given

    Discriminative models take much longer to train than their generative counterpart.

    “First, discriminative models take much longer to train than their generative counterpart.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationnone given

    HSMM results in slower inference and is dependent on correct modelling assumptions for the durations.

    “It does, however, result in slower inference and is dependent on correct modelling assumptions for the durations.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Truyen applied a hierarchical version of SMCRFs to activity recognition using a small laboratory dataset, and the hierarchical SMCRF outperformed both hierarchical CRF and conventional CRF.

    “Truyen applied a hierarchical version of SMCRFs to activity recognition using a small dataset recorded in a laboratory setting [24].The performance was compared to hierarchical CRF and a conventional CRF, the hierarchical SMCRF outperformed both.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [24] 24. T.T. Truyen, H.H. Bui, and S. Venkatesh. Human activity learning and segmentation using partially hidden discrimina- tive models. In HAREM, 2005.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Most early work on activity recognition used HMMs to recognize activities from sensor data.

    “Most early work on activity recognition used HMMs to recognize the activities from sensor data [14,28].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [14] 14. D.L. Vail. Conditional Random Fields for Activity Recogni- tion. PhD Thesis, Carnegie Mellon University, 2008.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [28] 28. H. Wallach. Efficient training of conditional random fields. Master's thesis, University of Edinburgh, 2002.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation

    State durations in HMMs and CRFs are modelled implicitly by means of self-transitions of states, which entails important limitations.

    “State durations are modelled implicitly by means of self-transitions of states, and this entails a number of important limita-tions [10,15].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [10] 10. E. Marhasev, M. Hadad, and G.A. Kaminka. Non-stationary hidden semi Markov models in activity recognition. In MOO, 2006.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [15] 15. D.J. Patterson, D. Fox, H.A. Kautz, and M. Philipose. Fine- grained activity recognition by aggregating abstract object us- age. In ISWC, pages 44-51. IEEE Computer Society, 2005. source

  • May combine two claims.

    Study design and conductown data

    The training data are fully labelled with exact start and end time of each segment, so that parameters can be optimized in closed form for both HMMs and HSMMs.

    “Our training data are fully labelled, with exact start and end time of each segment, so that we can optimize the parameters in closed form for both HMMs and HSMMs.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    Hierarchical models require a lot more parameters proportional to the number of substates.

    “Although this allows better duration modelling than conventional models, it also requires a lot more parameters (proportional to the number of substates).”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisenone given

    HMMs only model the observations and transitions between hidden states, while HSMMs also model the duration of hidden states explicitly.

    “They differ in the sense that HMMs only model the observations and transitions between hidden states, while HSMMs also model the duration of hidden states explicitly.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Modelling the data more accurately using the HSMM allows both speedy learning and good performance, and is less prone to overfitting.

    “Modelling the data more accurately using the HSMM allows both speedy learning and good performance, and is less prone to overfitting.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    After CRFs had been shown to outperform HMMs in other areas, they were applied to activity recognition as well.

    “After CRFs had been shown to outperform HMMs in other areas [6], they were applied to activity recognition as well [25,26].”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [6] 6. J. Lafferty, A. McCallum, and F. Pereira. Conditional ran- dom fields: Probabilistic models for segmenting and labeling sequence data. In Proc. 18th International Conf. on Machine Learning, pages 282-289.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [25] 25. T.T. Truyen, D.Q. Phung, H.H. Bui, and S. Venkatesh. Hi- erarchical semi-Markov conditional random fields for recur- sive sequential data. In Neural Information Processing Systems (NIPS), 2008.

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [26] 26. D.L. Vail, M.M. Veloso, and J.D. Lafferty. Conditional random fields for activity recognition. In International Conference on Autonomous Agents and Multi-agent Systems (AAMAS), 2007. source

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Interpretationreasoning

    Because certain durations are very unlikely for activities such as toileting and showering, the 'other activity' is the only activity left because it can practically have any duration.

    “Because certain durations are very unlikely for activities such as 'toiletting' and 'showering', the 'other activity' is the only activity left because it can practically have any duration.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    Rabiner pointed out that modelling of duration was one of the major weaknesses of HMMs and suggested the use of HSMMs.

    “Rabiner had already pointed out that the modelling of duration was one of the major weaknesses of HMMs and suggested the use of HSMMs [15].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [15] 15. D.J. Patterson, D. Fox, H.A. Kautz, and M. Philipose. Fine- grained activity recognition by aggregating abstract object us- age. In ISWC, pages 44-51. IEEE Computer Society, 2005. source
18 of 18

  • The authors state this as an assumption.

    The observations used in the models are binary.

    “We assume our observations are binary, for convenience, and because the datasets used for evaluation consist only of binary observations.”

  • The paper relies on this without support.

    In the application, activities might be repeated shortly after each other, so it makes more sense to model the data with self-transitions.

    “HSMMs have been proposed before where such a constraint was enforced [15], but in our application activities might be repeated shortly after each other (e.g., getting another drink), so that it makes more sense to model the data with self-transitions.”

  • The paper relies on this without support.

    Shaving typically takes up more time than brushing teeth.

    “However, because shaving typically takes up more time than brushing teeth, the duration of the activity is likely to be very informative for recognition.”

  • The paper relies on this without support.

    Installing extra sensors to further distinguish between activities is not always possible.

    “Installing extra sensors to further distinguish between activities is not always possible because it is difficult to install a sensor on, for example, a tooth brush or to sense which item is retrieved from a refrigerator.”

  • The paper relies on this without support.

    There is a strong need for real-world datasets in the activity recognition community.

    “There is a strong need for real world datasets in the activity recognition community.”

  • The paper relies on this without support.

    The 'other' activity is a collection of activities such as being idle, not a single activity type.

    “The 'other' activity is not one type of activity, but rather a collection of activities such as 'being idle'.”

  • The paper relies on this without support.

    The activity annotations in the datasets are accurate ground truth for the activities performed.

    “Annotation for the house 1 datasets was recorded using a bluetooth headset combined with speech recognition.”
    “In house 2 a handwritten activity diary was used for annotation.”
7 of 7

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Transition estimate uses the wrong conditioning-state count

    Minor

    The stated transition probability is conditioned on the previous state being i, but the displayed estimate divides by the count of previous states equal to j. Its denominator therefore does not match the probability it purports to estimate.

  • SMCRF does not improve recognition of every activity

    Minor

    The paper says SMCRF performs better than CRF for every activity, but both confusion matrices show zero correctly recognized brush timeslices.

2 of 2

Cites

Flagged first, then by citations

5 of the 35 references couldn’t be matched to a known paper.

Research works printed but not linked: 3

  1. [1]1. G. Abowd, A. Bobick, I. Essa, E. Mynatt, and W. Rogers. The aware home: Developing technologies for successful ag- ing. In Proceedings of AAAI Workshop and Automation as a Care Giver, 2002.
  2. [7]7. Morgan Kaufmann, San Francisco, CA, 2001.
  3. [14]14. D.L. Vail. Conditional Random Fields for Activity Recogni- tion. PhD Thesis, Carnegie Mellon University, 2008.

Cited by

Sorted by citations

Related

Authors
van Kasteren, T. L. M., Englebienne, G., & Kröse, B. J. A.
Affiliations
University of Amsterdam (NL)
ORCID
Venue
Journal of Ambient Intelligence and Smart Environments
Publisher
SAGE Publications
Published
2010
Volume
2
Issue
3
Pages
311-325
ISSN
1876-1364 (ISSN-L)
DOI
10.3233/ais-2010-0070
OpenAlex
W2126484555
Semantic Scholar
a1a098a6b2acb1c9e098a7681d3535d9860f3c52
Cited by
131
References
35
Citations by year
2026: 2 · 2025: 1 · 2024: 6 · 2023: 6 · 2022: 7 · 2021: 10 · 2020: 5 · 2019: 11 · 2018: 9 · 2017: 11 · 2016: 13 · 2015: 16 · 2014: 13 · 2013: 10 · 2012: 7
FWCI
7.79
Type
journal article
Language
en
Open access
Locations
  • Journal of Ambient Intelligence and Smart Environments (publishedVersion)
  • UvA-DARE (University of Amsterdam) (publishedVersion, other-oa)
  • UvA-DARE (University of Amsterdam) (submittedVersion, other-oa)
  • Data Archiving and Networked Services (DANS) (submittedVersion, other-oa)
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Context-Aware Activity Recognition Systems
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Human Pose and Action Recognition
  • Social Sciences › Transportation › Human Mobility and Location-Based Analysis
Keywords
activity recognition, semi-Markov model, home automation, sensor data, hidden Markov models, conditional random fields, probabilistic sequence models, human activity recognition

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-11.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-11.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review