On-body activity recognition systems are becoming more and more frequent in people's lives. These systems normally register body motion signals through small sensors that are placed on the user. To perform the activity detection the signals must be adequately partitioned, however, no clear consensus exists on how this should be done. More specically, considered the sliding window technique the most widely used approach for segmentation, it is unclear which window size must be applied. This presentation investigates the effects of the windowing procedure on the activity recognition process. To that end, diverse recognition systems are tested for several window sizes also including the figures used in previous works. From the study it may be concluded that reduced window sizes lead to a better recognition of the activities, which goes against the generalized idea of using long data windows. This presentation illustrates part of the work described in the following articles: * Banos, O., Galvez, J. M., Damas, M., Pomares, H., Rojas, I. Window size impact in activity recognition. Sensors, vol. 14, no. 4, pp. 6474-6499 (2014) * Banos, O., Galvez, J. M., Damas, M., Pomares, H., Rojas, I.: Evaluating the effects of signal segmentation on activity recognition. In: Proceedings of the International Work-Conference on Bioinformatics and Biomedical Engineering (IWBBIO 2014), Granada, Spain, April 7-9, (2014)

1. Evaluating the effects of
signal segmentation on
activity recognition
IWBBIO 2014, Granada (España)
Oresti Baños, J.M. Gálvez, M. Damas, A. Guillén, L. J. Herrera, H. Pomares, and I. Rojas
Department of Computer Architecture and Computer Technology,
Research Center for Information and Communications Technologies of the
University of Granada (CITIC-UGR), SPAIN
oresti@ugr.es

2. Introduction
• Activity recognition concept
– “Recognize the actions and goals of one or more agents from a series of
observations on the agents' actions and the environmental conditions”
• Applications (among others)
– eHealth (AAL, telerehabilation)
– Sports (performance improvement, injury-free pose)
– Industrial (assembly tasks, avoidance of risk situations)
– Gaming (Kinect, Wii Mote, PlayStationMove)
• Categorization by sensor modality
– Ambient (cameras, microphones, RFID)
– On-body (wearables)
2

3. Activity Recognition Chain (ARC)
3

4. Activity Recognition Chain (ARC)
4

5. Activity Recognition Chain (ARC)
5

6. Activity Recognition Chain (ARC)
6

7. Activity Recognition Chain (ARC)
7

8. Activity Recognition Chain (ARC)
8

9. Activity Recognition Chain (ARC)
9

10. Activity Recognition Chain (ARC)
10

11. Segmentation
• Types
11
– Simplest (no preprocessing)
– Most widely-used
– Window sizes (WS) ranges
from 0.1s to 15s and more
Sliding window Event-basedActivity-defined
– Analysis of activity changes
(spotting)
– Limitedly used
– Identification of
characteristic
events
– Mainly used in gait
analysis

12. Segmentation
• Types
12
– Simplest (no preprocessing)
– Most widely-used
– Window sizes (WS) ranges
from 0.1s to 15s and more
Sliding window Event-basedActivity-defined
But… which WS
should we use? No consensus!!!
A study is lacking!!!

13. Experimental setup: dataset
• Fitness benchmark dataset
• 33 activities
• 9 IMUs (XSENS)  ACC, GYR, MAG
• 17 subjects
13
Baños, O., Toth M. A., Damas, M., Pomares, H., Rojas, I., Amft, O.: A benchmark dataset to evaluate sensor displacement in activity recognition.
In: 14th International Conference on Ubiquitous Computing (Ubicomp 2012), Pittsburgh, USA, September 5-8, (2012)

14. Results
14
• Preprocessing: NO ● Segmentation: 0.25s-7s in steps of 0.25s
• Feature extraction: FS1={mean}, FS2={mean,std}, FS3={mean,std,max,min,mcr}
• Classification: DT, NB, NCC, KNN ● Evaluation: 10-fold cross-validation, 100 repetitions
Experimental Parameters

15. Conclusions and final remarks
• Segmentation is a crucial stage in activity recognition, however, there is no
clear consensus on how to partitionate the sensor data stream
• Sliding window is the most widely-used segmentation technique, but
there is no study that neatly investigates the impact of the window size
• We have performed an extensive study for various standard activity
recognition models evaluated for a wide range of window sizes and
activities
• Short windows (1-2s) provide the most accurate detection performance,
thus proving that using large window sizes does not necessarily translate
into a better recognition performance
• An extension of this study* provides designers with a set of practical
guidelines for the windowing process and for diverse activity categories
and applications
* Banos, O., Galvez, J. M., Damas, M., Pomares, H., Rojas, I. Window size impact in activity recognition. SENSORS. (2014) 15

16. Thank you for your attention.
Questions?
Oresti Baños Legrán
Department of Computer Architecture and Computer Technology,
Research Center for Information and Communications Technologies of the
University of Granada (CITIC-UGR)
Email: oresti@ugr.es
Web: http://www.ugr.es/~oresti
Phone: +34 958 241 778
Fax: +34 958 248 993
This work was partially supported by the Spanish CICYT Project SAF2010-20558, Junta de Andalucia Project P09-TIC-175476 and the FPU
Spanish grant AP2009-2244. 16