WISDM Related Work

General Research Papers


Biometrics

  • J. Frank, S. Mannor, and D. Precup, Activity and Gait Recognition with Time-Delay Embeddings. In Proceedings of the 24th AAAI Conference on Artificial Intelligence, Atlanta, GA, 2010.
    There is a related clip on the Discovery channel (the clip starts midway through the segment).
  • R. Srinivasan, C. Chen, and D. Cook (2010). Activity Recognition using Actigraph Sensor. In Proceedings of the 4th International workshop on Knowledge Discovery from Sensor Data, 2010.
  • D. Gafurov and E. Snekkenes (2009). Gait Recognition Using Wearable Motion Recording Sensors. In EURASIP Journal on Advances in Signal Processing, 2009: Article ID 415817.

    Researchers analyzed data collected from motion recording sensors with tri-axial accelerometers placed on the foot, hip, pocket, and arm (in separate experiments). They examined the best performances of recognition methods based on the motion of these different body parts, as well as how robust gait-based authentication is under three attack scenarios and what attributes contribute to the uniqueness of human gait.


  • D. Gafurov (2007). A Survey of Biometric Gait Recognition: Approaches, Security, and Challenges. In Annual Norwegian Computer Science Conference, November 19-21, 2007.

    Gafurov presents an overview of recent biometric research focusing on gait recognition. He first describes basic evaluation metrics in biometrics, including false accept rates, false reject rates, equal error rates, DET curves, and CMC curves. He identifies three types of gait recognition--machine vision based, floor sensor based, and wearable sensor based--and describes various challenges that the wearable sensor based-approach encounters.


  • D. Gafurov, K. Helkala and T. Sondrol (2006). Biometric Gait Authentication Using Accelerometer Sensor. In Journal of Computers, 1(7):51-59.

    This paper examines how gait patterns can be used as an unobtrusive means to authenticate a user’s identity. Gait patterns were recorded by attaching an accelerometer to the user’s right lower leg. Histogram similarity and cycle length were used as part of the authentication procedure, which was tested on a population of 21 participants.


  • J. Mantyjarvi, M. Lindholdm,et. al. (2005). Identifying Users of Portable Devices from Gait Pattern with Accelerometers. In Proceedings of IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP '05).

    The goal of this work was to identify users using data from accelerometers in portable devices that users normally carry. Data was collected from 36 subjects who walked at fast, normal, and slow walking speeds on two separate days. The accelerometer device was placed on their belts, at the middle of their waistline in back. Correlation, frequency domain, and data distribution statistics were used to identify users.



Activity Monitoring

  • J. Yang (2009). Toward physical activity diary: motion recognition using simple acceleration features with mobile phones. In ICME '09 Proceedings of the 1st International Workshop on Interactive Multimedia for Consumer Electronics.
  • T. Brezmes, J.L. Gorricho and J. Cotrina (2009). Activity Recognition from Accelerometer Data on Mobile Phones. In IWANN '09: Proceedings of the 10th International Work-Conference on Artificial Neural Networks, 796-799.

    Researchers implemented a realtime system for classifying six basic daily activities using a mobile phone containing an accelerometer. No server processing data was involved. The K-nearest neighbors algorithm was used, with the intent that users could train the device to detect their motions for whatever location the particular user normally carries his or her phone.


  • N. Gyorbiro, A. Fabian and G. Homanyi (2008). An Activity Recognition System for Mobile Phones. In Mobile Networks and Applications , 14:82-91.

    Researchers used feed-forward backpropogation neural networks to distinguish between six different motion patterns using data collected from three MotionBands attached to the dominant wrist, hip, and ankle of each subject. Each MotionBand contained a tri-axial accelerometer, magnetometer, and gyroscope. A smartphone collected data from the MotionBand sensors.


  • N. Krishnan, D. Colbry, et. al. (2008) Real Time Human Activity Recognition Using Tri-Axial Accelerometers. In Sensors Signals and Information Processing Workshop.

    Researchers designed a real time system for identifying five lower body activities with data from 3 subjects collected from tri-axial accelerometers. Accelerometers were placed on the right ankle and the left thigh. Researchers extracted statistical and spectral features from the data, including mean, variance, energy, spectral entropy, and correlation between the data of all axes, and the AdaBoost algorithm built on decision stump for classification was trained with three-fold cross validation, and, in addition, probability of classifications were calculated.


  • N. Krishnan and S. Panchanathan (2008). “Analysis of Low Resolution Accelerometer Data for Continuous Human Activity Recognition.” In IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2008), 3337-3340.

    Researchers evaluated the performance of different classifiers on a subset of the accelerometer data used in the Bao and Intille experiment. Classifiers included AdaBoost, SVM, and Regularized Logistic Regression (RLogReg). Ten random subjects and seven lower body activities were examined. The continuous acceleration stream was divided into fixed length frames, and each frame was classified. Statistical features including mean, variance, correlation between all the axis of the accelerometers and spectral features including energy and entropy were calculated.


  • Y. Cho, Y. Nam, et. al. (2008). SmartBuckle: Human Activity Recognition using a 3-axis Accelerometer and a Wearable Camera. In HealthNet, ’08.

    Researchers used the SmartBuckle device for medical monitoring to recognize each of 9 different activities. The device contained a tri-axial accelerometer as well as an image sensor. Correlation between axes and the magnitude of the FFT were used as features for the accelerometer data.


  • E.M. Tapia, S.S. Intille, et al. (2007). Real-Time Recognition of Physical Activities and Their Intensities Using Wireless Accelerometers and a Heart Rate Monitor . In Proceedings of the 2007 11th IEEE International Symposium on Wearable Computers, 1-4.

    Researchers used five triaxial accelerometers to collect data in real-time from 21 users while the users performed thirty different gymnasium activities. Several of these "activities" involved performing the same activity at different levels of intensity. Using C4.5 DT and the Naive Bayes classifiers in WEKA, researchers found that they could achieve high accuracies of activity recognition for subject dependent analysis but much lower accuracies for subject independent analysis. Adding in a heart rate monitor only slightly improved the results, and combining activities that were the same activity being performed at different intensities also improved accuracies slightly more.


  • U. Maurer, A. Smailagic, et. al. (2006). Activity Recognition and Monitoring Using Multiple Sensors on Different Body Positions. In IEEE Proceedings of the International Workshop on Wearable and Implantable Body Sensor Networks,3(5):113-116.

    Researchers sought to identify users’ activities in realtime using “eWatch” sensors placed on the belt, shirt pocket, trouser pocket, backpack, and necklace. Features from the accelerometer axes, the light sensor, and a combined value of the accelerometer signals were calculated. The Correlation-based Feature Selection (CFS) method from WEKA was used to find feature sets that were highly correlated with a particular class but uncorrelated with each other.


  • N. Ravi, N. Dandekar, et. al. (2005). Activity Recognition from Accelerometer Data. In Proceedings of the Seventeenth Conference on Innovative Applications of Artificial Intelligence.

    Researchers collected data from a triaxial accelerometer worn at subjects' waists. Mean, standard deviation, energy, and correlation features were extracted from the data. In addition to analyzing the performance of base-level classifiers like decision tables, decision trees, K-nearest neighbors, SVM, and Naive Bayes, meta-level classifiers such as boosting, bagging, plurality voting, stacking using ODTs, and stacking using MDTs were applied to classify windows as one of eight daily activities.


  • L. Bao and S. Intille (2004). Activity Recognition from User-Annotated Acceleration Data . In PERVASIVE, LNCS 3001, 1–17.

    Five biaxial accelerometers placed on the right hip, dominant wrist, non-dominant upper arm, dominant ankle, and non-dominant thigh were used to collect data from 20 subjects. Twenty daily activities were considered. Mean energy, frequency-domain entropy, and correlation of acceleration data were calculated, and C4.5, instance-based learning, decision tables, and Naive Bayes classifiers in WEKA were tested using these features.


  • M. Mathie, B. Celler et. al. (2004). Classification of Basic Daily Movements Using a Triaxial Accelerometer. In Medical & Biological Engineering & Computing, 42:679-687.

    Researchers developed their own binary tree for classification of basic daily activities and created algorithms to describe each of these activities. Using this model, they tested data collected from 26 subjects using a triaxial accelerometer attached at the subjects’ waists. Of particular interest in this study was the detection of falls.



Geo-spatial Data Mining (of GPS Data)


Related Research Groups and Stories