Haar-Like Filtering for Human Activity Recognition Using 3D Accelerometer

In this paper, novel 2 one-dimensional (1D) Haar-like filtering techniques are proposed as a new and low calculation cost feature extraction method suitable for 3D acceleration signals based human activity recognition. Proposed filtering method is a simple difference filter with variable filter parameters. Our method holds a strong adaptability to various classification problems which no previously studied features (mean, standard deviation, etc.) possessed. In our experiment on human activity recognition, the proposed method achieved both the highest recognition accuracy of 93.91% while reducing calculation cost to 21.22% compared to previous method.

[1]  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.

[2]  Patrik Floréen,et al.  A Framework for Distributed Activity Recognition in Ubiquitous Systems , 2005, IC-AI.

[3]  Michael L. Littman,et al.  Activity Recognition from Accelerometer Data , 2005, AAAI.

[4]  Bernt Schiele,et al.  Analyzing features for activity recognition , 2005, sOc-EUSAI '05.

[5]  Tadahiro Kuroda,et al.  Speaker Siglet Detection for Business Microscope , 2008, 2008 Seventh International Conference on Machine Learning and Applications.

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

[7]  T. Kuroda,et al.  Low cost speech detection using Haar-like filtering for sensornet , 2008, 2008 9th International Conference on Signal Processing.

[8]  Tadahiro Kuroda,et al.  Speech "Siglet" Detection for Business Microscope (concise contribution) , 2008, 2008 Sixth Annual IEEE International Conference on Pervasive Computing and Communications (PerCom).