So far little attention has been paid to activity recognition systems limitations during out-of-lab daily usage. Sensor displacement is one of these major issues, particularly deleterious for inertial on-body sensing. The effect of the displacement normally translates into a drift on the signal space that further propagates to the feature level, thus modifying the expected behavior of the predened recognition systems. On the use of several sensors and diverse motion-sensing modalities, in this paper we compare two fusion methods to evaluate the importance
of decoupling the combination process at feature and classication levels under realistic sensor congurations. In particular a 'feature fusion' and a 'multi-sensor hierarchical-classifier' are considered. The results reveal that the aggregation of sensor-based decisions may overcome the
difficulties introduced by the displacement and confirm the gyroscope as possibly the most displacement-robust sensor modality.
This presentation illustrates part of the work described in the following articles:
* Banos, O., Damas, M., Pomares, H., Rojas, I.: Handling displacement effects in on-body sensor-based activity recognition. In: Proceedings of the 5th International Work-conference on Ambient Assisted Living an Active Ageing (IWAAL 2013), San José, Costa Rica, December 2-6, (2013)
* Banos, O., Damas, M., Pomares, H., Rojas, I. On the Use of Sensor Fusion to Reduce the Impact of Rotational and Additive Noise in Human Activity Recognition. Sensors, vol. 12, no. 6, pp. 8039-8054 (2012)
Handling displacement effects in on-body sensor-based activity recognition
1. Handling displacement
effects in on-body sensor-
based activity recognition
IWAAL 2013, San José (Costa Rica)
Oresti Baños, Miguel Damas, Héctor Pomares, and Ignacio Rojas
Department of Computer Architecture and Computer Technology, CITIC-UGR,
University of Granada, SPAIN
oresti@ugr.es
DG-Research Grant #228398
3. Context
• On-body activity recognition is becoming true… Really?
– Reliability
• Different performance depending on who uses the system (age,
height, gender,…) and due to people changes during the lifelong use
(conditions, ageing,…) Reliable? Perdurable?
– Usability/Applicability
• Application-specific systems require to put on several diverse
systems to provide different functionalities Portable?
Unobtrusive? Fashionable? Tractable?
– Robustness
• Sensor anomalies (decalibration, loose of attachment,
displacement,…) Robust?
3
7. Concept of sensor displacement
• Categories of sensor displacement
– Static: position changes can remain static across the execution of many activity
instances, e.g. when sensors are attached with a displacement each day
– Dynamic: effect of loose fitting of the sensors, e.g. when attached to cloths
• Sensor displacement new sensor position signal space change
• Sensor displacement effect depends on
– Original/end position and body part
– Activity/gestures/movements performed
– Sensor modality (ACC, GYR, MAG)
7
Sensor displacement = rotation + translation
(angular displacement) (linear displacement)
8. Sensor displacement effects
Changes in the signal
space forward
propagates on the
activity recognition
process (e.g., variations
in the feature space)
RCIDEAL LCIDEAL= LCSELF
8
RCSELF
11. Multi-Sensor Hierarchical Classifier
11
SM
S2
S1
α11
∑
C12
C1N
C11
∑
C21
C22
C2N
∑
CM1
CM2
CMN
∑
Decision
Class level Source level Fusion
β11
α12
β12
α1N
β1N
α21
β21
α22
β22
α2N
β2N
αM1
βM1
αM2
βM2
αMN
βMN
γ11,…,1N
δ11,…,1N
γ21,…,2N
δ21,…,2N
γM1,…,MN
δM1,…,MN
[-0.14,3.41,4,21,…,6.11]
[-0.84,3.21,4.21,…,6.11]
[-0.81,5.71,4.21,…,6.22]
[-0.14,3.92,4.23,…,7.82]
S1
S2
SM
u1 p1 s11,s12,…,s1k fℝ(s11,s12,…,s1k)
u2 p2 s21,s22,…,s2k fℝ(s21,s22,…,s2k)
uM pM sM1,sM2,…,sMk fℝ(sM1,sM2,…,sMk)
O. Banos, M. Damas, H. Pomares, F. Rojas, B. Delgado-Marquez, and O. Valenzuela. Human activity recognition
based on a sensor weighting hierarchical classifier. Soft Computing, 17:333-343, 2013.
12. Study of sensor displacement effects
• Analyze
– Variability introduced by sensors self-positioning with respect to an ideal
setup
– Effects of large sensor displacements (extreme de-positioning)
– Robustness of sensor fusion to displacement
• Scenarios
– Ideal-placement
– Self-placement
– Induced-displacement
Ideal Self Induced
12
O. Banos, M. Damas, H. Pomares, I. Rojas, M. Attila Toth, and O. Amft.
A benchmark dataset to evaluate sensor displacement in activity
recognition. In Proceedings of the 2012 ACM Conference on
Ubiquitous Computing, pages 1026-1035, New York, NY, USA, 2012.
14. Dataset: Study setup
• Cardio-fitness room
• 9 IMUs (XSENS) ACC, GYR, MAG
• Laptop data storage and labeling
• Camera offline data validation
http://crnt.sourceforge.net/CRN_Toolbox/Home.html14
15. Dataset: Experimental protocol
• Scenario description
• Protocol
Round Sensor Deployment #subjects #anomalous
sensors
1st Self-placement 17 3/9
2nd Ideal-placement 17 0/9
- Mutual-displacement 3 {4,5,6 or 7}/9
15
Preparation phase
(sensor positioning &
wiring, Xsens-Laptop
bluetooth connection,
camera set up)
Exercises execution
(20 times/1 min. each)
Battery replacement,
data downloading
Data postprocessing
(relabeling, visual
inspection,
evaluation)
Round
16. Experimental setup
• Data considerations
– Data domain: ACC, GYR, MAG
and combinations (ACC-GYR,
ACC-MAG, GYR-MAG,
ACC-GYR-MAG)
– ALL sensors
• Activity recognition methods
– No preprocessing
– Segmentation: 6 seconds
sliding window
– Features: MEAN, STD, MAX,
MIN, MCR
– Reasoner: C4.5 decision tree
16
• Sensor displacement scenarios
– Ideal (no displacement)
– Self (3 out of all sensors)
– Induced (7 out of all sensors)
• Evaluation
– Ideal: 5-fold cross validation,
100 times
– Self/Mutual: tested on a
system trained on ideal-
placement data
20. Conclusions and final remarks
• Sensor anomalies (here displacement) may seriously damage the
performance of activity recognition systems, especially single sensor
based systems
• Sensor fusion is proposed to deal with these anomalies
• Feature fusion approaches (the most widely used) has been
demonstrated to be very sensitive to sensor displacement
• Decision fusion cope much better with the effects of sensor
displacement, even when a majority of the sensors are highly de-
positioned
• From the analyzed sensor magnitudes, GYR outstands as the most
robust modality to displacement with a performance drop of less than
5% for the self-placement scenario and 10% for the extreme
displacement 20
21. Thank you for your attention.
Questions?
Oresti Baños Legrán
Dep. Computer Architecture & Computer Technology
Faculty of Computer & Electrical Engineering (ETSIIT)
University of Granada, Granada (SPAIN)
Email: oresti@ugr.es
Phone: +34 958 241 778
Fax: +34 958 248 993
Work supported in part by the HPC-Europa2 project funded by the European Commission - DG Research in the Seventh Framework Programme
under grant agreement no. 228398 and the FPU Spanish grant AP2009-2244. 21