Presentazione human daily activity recognition with sparse representation using wearable sensors
-
Upload
fabio-greco -
Category
Healthcare
-
view
128 -
download
0
Transcript of Presentazione human daily activity recognition with sparse representation using wearable sensors
Case Study :“Human Daily Activity Recognition With Sparse
Representation Using Wearable Sensors”IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS, VOL.17, NO. 3, MAY 2013 Mi Zhang, Student Member, IEEE, and Alexander A. Sawchuk, Life Fellow, IEEE
Giuseppe Gagliano , Fabio Greco Pisa, 24/05/2016
Introduction on Pervasive Healthcare :
who and why ?
Among all the human activity sensing technologies, wearable sensing system has the advantage of being with people throughout the day, enabling continuously collecting people’s activity information.
for what ?
To deliver long-term personalized fitness monitoring, preventive and chronic healthcare, and elderly support.
how ?
This is a pattern recognition problem so...
Feature Extraction
● Which features do I extract?
● Should I select features or should I use Random Projection?
Training
Feature Extraction
● Which features do I extract?
● Should I select features or should I use Random Projection?
Overcomplete Dictionary Construction & Sparse Representation
● Feature Vectors
Training
Feature Extraction
● Which features do I extract?
● Should I select features or should I use Random Projection?
Overcomplete Dictionary Construction & Sparse Representation
● Feature Vectors
Training
D1 Dk
Feature Extraction
● Which features do I extract?
● Should I select features or should I use Random Projection?
Overcomplete Dictionary Construction & Sparse Representation
● Feature Vectors
● Overcomplete Dictionary
● Sparse coefficient
Training
D1 Dk
Recognition
1. Sparse recovery via l1 minimization
2. Classification via Sparse Representation (SR)
3. Sparsity confidence measure
Recognition
1. Sparse recovery via l1 minimization
2. Classification via Sparse Representation (SR)
3. Sparsity confidence measure
Effect of the Feature Dimension and Comparison to Baseline Algorithms
NN - Nearest Neighbour
NBC - Naive Bayesian Classifier
SVM - Support Vector Machine
SR - Sparse Representation