Recognizing Soldier Activities in the Field

We describe the activity recognition component of the Soldier Assist System (SAS), which was built to meet the goals of DARPA’s Advanced Soldier Sensor Information System and Technology (ASSIST) program. As a whole, SAS provides an integrated solution that includes on-body data capture, automatic recognition of soldier activity, and a multimedia interface that combines data search and exploration. The recognition component analyzes readings from six on-body accelerometers to identify activity. The activities are modeled by boosted 1D classifiers, which allows efficient selection of the most useful features within the learning algorithm. We present empirical results based on data collected at Georgia Tech and at the Army’s Aberdeen Proving Grounds during official testing by a DARPA appointed NIST evaluation team. Our approach achieves 78.7% for continuous event recognition and 70.3% frame level accuracy. The accuracy increases to 90.3% and 90.3% respectively when considering only the modeled activities. In addition to standard error metrics, we discuss error division diagrams (EDDs) for several Aberdeen data sequences to provide a rich visual representation of the performance of our system.

[1]  Paul Lukowicz,et al.  Performance Metrics and Evaluation Issues for Continuous Activity Recognition , 2006 .

[2]  Gregory D. Abowd,et al.  Recognizing mimicked autistic self-stimulatory behaviors using HMMs , 2005, Ninth IEEE International Symposium on Wearable Computers (ISWC'05).

[3]  Paul A. Viola,et al.  Rapid object detection using a boosted cascade of simple features , 2001, Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. CVPR 2001.

[4]  Blake Hannaford,et al.  A Hybrid Discriminative/Generative Approach for Modeling Human Activities , 2005, IJCAI.

[5]  Robert E. Schapire,et al.  A Brief Introduction to Boosting , 1999, IJCAI.

[6]  Ling Bao,et al.  Activity Recognition from User-Annotated Acceleration Data , 2004, Pervasive.

[7]  Bernt Schiele,et al.  A model for human interruptability: experimental evaluation and automatic estimation from wearable sensors , 2004, Eighth International Symposium on Wearable Computers.

[8]  Paul Lukowicz,et al.  Evaluating Performance in Continuous Context Recognition Using Event-Driven Error Characterisation , 2006, LoCA.

[9]  Jeff A. Bilmes,et al.  Rao-Blackwellized Particle Filters for Recognizing Activities and Spatial Context from Wearable Sensors , 2006, ISER.

[10]  John Platt,et al.  Probabilistic Outputs for Support vector Machines and Comparisons to Regularized Likelihood Methods , 1999 .

[11]  Paul Lukowicz,et al.  SoundButton: design of a low power wearable audio classification system , 2003, Seventh IEEE International Symposium on Wearable Computers, 2003. Proceedings..

[12]  Paul Lukowicz,et al.  Recognizing Workshop Activity Using Body Worn Microphones and Accelerometers , 2004, Pervasive.