There exist multiple activity recognition solutions offering
good results under controlled conditions. However, little attention has been given to the development of functional systems operating in realistic settings. In that vein, this work aims at presenting the complete process for the design, implementation and evaluation of a real-time activity recognition system. The proposed recognition system consists of three wearable inertial sensors used to register the user body motion, and a mobile application to collect and process the sensory data for the recognition of the user activity. The system not only shows good recognition capabilities after online evaluation but also after analysis at runtime. In view of the obtained results, this system may serve for the recognition
of some of the most frequent daily physical activities.
On the Development of A Real-Time Multi-Sensor Activity Recognition System
1. On the Development of
A Real-Time Multi-Sensor
Activity Recognition System
IWAAL 2015 (Puerto Varas, Chile)
Oresti Banos1, Miguel Damas2, Alberto Guillen2, Luis-Javier Herrera2, Hector Pomares2,
Ignacio Rojas2, Claudia Villalonga1,2 and Sungyong Lee1
1Ubiquitous Computing Lab, Kyung Hee University, Korea
2Department of Computer Architecture and Computer Technology, University of Granada, Spain
2. Wearable activity recognition
The Activity
Recognition
Chain (ARC)
Phenomena
Human activity
(body motion)
Measurement
Sensing
(wearables/on-body)
Processing
Data curation
and knowledge
inference Recognized
Activity
http://orestibanos.com/
3. Roadmap for the development of an online WAR system
Problem
specification
Activities, Sensors
and Processing
Hub selection
Dataset collection
Offline models
implementation
and evaluation
Online model
implementation
and evaluation
(Market…)
http://orestibanos.com/
5. Activities, Sensors and Processing
Single sensor vs Multiple sensor
Multi-sensorSingle sensor
IdealSelf
AR model/
#faulty sensors 0 1 2 3 4 5
New dynamic range= 30% original dynamic range [-3g,3g]
SARC (hip) 82±5 66±4 - - - -
SARC (wrist) 88±5 54±6 - - - -
SARC (arm) 80±3 58±7 - - - -
SARC (ankle) 83±4 58±8 - - - -
SARC (thigh) 89±2 72±4 - - - -
HWC 96±2 96±2 93±3 86±5 73±8 65±14
New dynamic range= 10% original dynamic range [-1g,1g]
SARC (hip) 82±5 21±11 - - - -
SARC (wrist) 88±5 18±9 - - - -
SARC (arm) 80±3 26±14 - - - -
SARC (ankle) 83±4 21±7 - - - -
SARC (thigh) 89±2 20±6 - - - -
HWC 96±2 94±2 87±6 53±2 27±17 25±19
Resiliency to Sensor Failures/Faults
Resiliency to Sensor Misplacements
Banos, O., Damas, M., Guillen, A., Herrera, L.J., Pomares, H., Rojas, I., Villalonga C. Multi-sensor fusion based on
asymmetric decision weighting for robust activity recognition. Neural Processing Letters, vol. 42, no. 1, pp. 5-26
vol. 42, no. 1, pp. 5-26 (2015).
Banos, O., Toth M. A., Damas, M., Pomares, H., Rojas, I. Dealing with the effects of sensor displacement in wearable
activity recognition. Sensors, vol. 14, no. 6, pp. 9995-10023 (2014).
http://orestibanos.com/
6. Activities, Sensors and Processing
Smartphone vs Cloud
Smartphone as a Platform Cloud as a Platform
Reasonable computing resources
X Limited storage (+ data is trapped into the device)
Widely available (“cheap”)
Data privacy is guaranteed while not distributed
to other platforms or services
“Unlimited” processing resources
“Unlimited” storage resources
X Expensive for small scale applications
X Complex management of data privacy
http://orestibanos.com/
8. Dataset collection
http://archive.ics.uci.edu/ml/
datasets/MHEALTH+Dataset
Sensing Modalities:
- 3D ACC, 3D GYR, 3D MAG, (+ 2-leads ECG)
Subjects:
- 10 males (27-45 years old)
Activities:
• Standing still (1 min)
• Sitting and relaxing (1 min)
• Lying down (1 min)
• Walking (1 min)
• Climbing/descending stairs (1 min)
• Waist bends forward (20x)
• Frontal elevation of arms (20x)
• Knees bending (crouching) (20x)
• Cycling (1 min)
• Jogging (1 min)
• Running (1 min)
• Jump front & back (20x)
MHEALTH Dataset
http://orestibanos.com/
9. Offline model implementation and evaluation
Parameters:
- Data
• 3D ACC from chest, wrist and ankle
- Segmentation
- Sliding window – 2sec (no-overlap)
- Feature set:
• FS1: mean + std
• FS2: FS1 + max + min + mcr
• FS3: FS2 + mode + median + kurtosis
- Classifier:
• NB (Naïve Bayes)
• DT (Decision Tree)
• NCC (Nearest Centroid Classifier)
• KNN (K-Nearest Neighbor)
Evaluation procedure:
- 10-fold CV
- 100 iterationsFS1 FS2 FS3
0.5
0.6
0.7
0.8
0.9
1
F-score
NB DT NCC KNN
http://orestibanos.com/
10. Offline model implementation and evaluation
Parameters:
- Data
• 3D ACC from chest, wrist and ankle
- Segmentation
- Sliding window – 2sec (no-overlap)
- Feature set:
• FS1: mean + std
• FS2: FS1 + max + min + mcr
• FS3: FS2 + mode + median + kurtosis
- Classifier:
• NB (Naïve Bayes)
• DT (Decision Tree)
• NCC (Nearest Centroid Classifier)
• KNN (K-Nearest Neighbor)
Evaluation procedure:
- 10-fold CV
- 100 iterationsFS1 FS2 FS3
0.5
0.6
0.7
0.8
0.9
1
F-score
NB DT NCC KNN
FS1 FS2 FS3
0.5
0.6
0.7
0.8
0.9
1
F-score
NB DT NCC KNN
FS1 FS2 FS3
0.5
0.6
0.7
0.8
0.9
1
F-score
NB DT NCC KNN
FS1 FS2 FS3
0.5
0.6
0.7
0.8
0.9
1
F-score
NB DT NCC KNN
http://orestibanos.com/
11. Online model implementation and evaluation
https://github.com/mHealthTech
nologies/mHealthDroid
Banos, O., Garcia, R., Holgado, J. A., Damas, M.,
Pomares, H., Rojas, I., Saez, A., Villalonga,
C. mHealthDroid: a novel framework for agile
development of mobile health applications. IWAAL
2014, Belfast, Northern Ireland, December 2-5, (2014).
Banos, O., Villalonga, C., Garcia, R., Saez, A., Damas,
J. A., Lee, S., Pomares, H., Rojas, I. Design,
validation of a novel open framework for agile
development of mobile health
applications. BioMedical Engineering OnLine, vol. 14,
14, no. S2:S6, pp. 1-20 (2015).
http://orestibanos.com/
15. Conclusions
• Most activity recognition works contribute with
models not validated in realistic conditions
• This work summarized the process for the
realization of a multi-sensor activity recognition
system oriented to real-time applications
• We especially contribute with an open
multimodal activity recognition dataset for the
benchmarking and development of new
recognition systems
• Next steps include to address identified barriers
to create activity recognition applications for
the real-world
http://orestibanos.com/
16. Work supported by the Junta de Andalucia Project P12-TIC-2082. This work was also supported by the ICTD Program (10049079, Development of mining core technology exploiting personal big data) funded by the
Ministry of Trade, Industry and Energy (MOTIE, Korea). We want to specially thank the participants who helped us to collect this dataset.
Thank you for your attention.
Questions?
Oresti Baños Legrán
Ubiquitous Computing Lab, Kyung Hee University, Korea
Email: oresti@oslab.khu.ac.kr
Web: http://orestibanos.com/
Hinweis der Redaktion
The idea of this work is to describe the process for developing a wearable AR system devised to operate in realistic settings and online