Analysis of Low Resolution Accelerometer Data for Continuous Human Activty Recognition. Krishnan, N. C. & Panchanathan, S.
Analysis of Low Resolution Accelerometer Data for Continuous Human Activty Recognition [link]Paper  doi  abstract   bibtex   
The advent of wearable sensors like accelerometers has opened a plethora\$\textbackslash{}backslash\$nof opportunities to recognize human activities from other low resolution\$\textbackslash{}backslash\$nsensory streams. In this paper we formulate recognizing activities from\$\textbackslash{}backslash\$naccelerometer data as a classification problem. In addition to the\$\textbackslash{}backslash\$nstatistical and spectral features extracted from the acceleration data,\$\textbackslash{}backslash\$nwe propose to extract features that characterize the variations in the\$\textbackslash{}backslash\$nfirst order derivative of the acceleration signal. We evaluate the\$\textbackslash{}backslash\$nperformance of different state of the art discriminative classifiers\$\textbackslash{}backslash\$nlike, boosted decision stumps (AdaBoost), support vector machines (SVM)\$\textbackslash{}backslash\$nand Regularized Logistic Regression(RLogReg) under three different\$\textbackslash{}backslash\$nevaluation scenarios(namely Subject Independent, Subject Adaptive and\$\textbackslash{}backslash\$nSubject Dependent). We propose a novel computationally inexpensive\$\textbackslash{}backslash\$nmethodology for incorporating smoothing classification temporally, that\$\textbackslash{}backslash\$ncan be coupled with any classifier with minimal training for classifying\$\textbackslash{}backslash\$ncontinuous sequences. While a 3% increase in the classification\$\textbackslash{}backslash\$naccuracy was observed on adding the new features, the proposed technique\$\textbackslash{}backslash\$nfor continuous recognition showed a 2.5-3% improvement in the\$\textbackslash{}backslash\$nperformance.
@article{krishnanAnalysisLowResolution2008,
  title = {Analysis of Low Resolution Accelerometer Data for Continuous Human Activty Recognition},
  issn = {15206149},
  url = {http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?arnumber=4518365},
  doi = {10.1109/ICASSP.2008.4518365},
  abstract = {The advent of wearable sensors like accelerometers has opened a plethora\$\textbackslash{}backslash\$nof opportunities to recognize human activities from other low resolution\$\textbackslash{}backslash\$nsensory streams. In this paper we formulate recognizing activities from\$\textbackslash{}backslash\$naccelerometer data as a classification problem. In addition to the\$\textbackslash{}backslash\$nstatistical and spectral features extracted from the acceleration data,\$\textbackslash{}backslash\$nwe propose to extract features that characterize the variations in the\$\textbackslash{}backslash\$nfirst order derivative of the acceleration signal. We evaluate the\$\textbackslash{}backslash\$nperformance of different state of the art discriminative classifiers\$\textbackslash{}backslash\$nlike, boosted decision stumps (AdaBoost), support vector machines (SVM)\$\textbackslash{}backslash\$nand Regularized Logistic Regression(RLogReg) under three different\$\textbackslash{}backslash\$nevaluation scenarios(namely Subject Independent, Subject Adaptive and\$\textbackslash{}backslash\$nSubject Dependent). We propose a novel computationally inexpensive\$\textbackslash{}backslash\$nmethodology for incorporating smoothing classification temporally, that\$\textbackslash{}backslash\$ncan be coupled with any classifier with minimal training for classifying\$\textbackslash{}backslash\$ncontinuous sequences. While a 3\% increase in the classification\$\textbackslash{}backslash\$naccuracy was observed on adding the new features, the proposed technique\$\textbackslash{}backslash\$nfor continuous recognition showed a 2.5-3\% improvement in the\$\textbackslash{}backslash\$nperformance.},
  journaltitle = {ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings},
  date = {2008},
  pages = {3337--3340},
  keywords = {Accelerometers,AdaBoost,Human activity recognition,SVM},
  author = {Krishnan, Narayanan C. and Panchanathan, Sethuraman},
  file = {/home/dimitri/Nextcloud/Zotero/storage/6LJCKWNP/Krishnan, Panchanathan - 2008 - Analysis of low resolution accelerometer data for continuous human activity recognition.pdf}
}
Downloads: 0