8-Pattern Recognition in Ubicom
-
Upload
dinh-ngoc-viet-tung -
Category
Documents
-
view
223 -
download
0
Transcript of 8-Pattern Recognition in Ubicom
-
8/6/2019 8-Pattern Recognition in Ubicom
1/29
Pattern Recognition inUbiquitous Computing
Moongu Jeon
GIST
-
8/6/2019 8-Pattern Recognition in Ubicom
2/29
Outline
Introduction to ubiquitous computing Mark Weiser The Computer for the Twenty-First Century"
Scientific American, pp. 94-10, September 1991.
Some Computer Science issues in UbiquitousComputing" Communications of the ACM, July1993.
Augmented reality Pattern recognition problems in Ubicomp
Introduction to Speech Recognition
-
8/6/2019 8-Pattern Recognition in Ubicom
3/29
Trend of Technology
Development
-
8/6/2019 8-Pattern Recognition in Ubicom
4/29
Ubiquitous Computing
Making computing an integral, invisible partof the way people live their lives, andavailable anytime anyplace.
Computers become parts of environment,and vanish into the background.
Integrating computers seamlessly into theworld.
-
8/6/2019 8-Pattern Recognition in Ubicom
5/29
Ubiquitous in IT
Writing The first IT freed information from the limits of
individual memory
Books,magazines, newspapers, street signs,billboards, shop signs, candy wrappers whichare parts of the environment- ubiquitous
Current silicon-based IT
Huge number of computers and communicationdevices is far from having becoming of theenvironment
-
8/6/2019 8-Pattern Recognition in Ubicom
6/29
Two Issues in Ubicomp
Location If a computer merely knows what room it is in,
it can adapt its behavior in significant ways
without requiring even a hint of artificialintelligence
Scale (size)
Tabs (inch-scale machine): Post-It notes
Pads (foot-scale): book or magazine
Boards (yard-scale): black (or bulletin) board
-
8/6/2019 8-Pattern Recognition in Ubicom
7/29
Tabs
ParcTab Olivetti Cambridge
Research Lab activebadges Identify and keep track
of users or objects, anddo more tasks.
Roy Want, PARC tab
incorporating a smalldisplay Serves simultaneously
as an active badge,calendar, diary
-
8/6/2019 8-Pattern Recognition in Ubicom
8/29
Pads
Scrap computer (analogous to scrappaper)
Can be grabbed and used anywhere.
Have no individual identity.
-
8/6/2019 8-Pattern Recognition in Ubicom
9/29
Boards
Large and shared display Video screen
Bulletin boards Whiteboards
Electronic bookcase
LiveBoard (PARC)
-
8/6/2019 8-Pattern Recognition in Ubicom
10/29
Other Issues Ubicomp
Cheap, low-power hardwarecomponents.
A network that ties them all together.
Software for screens and pens
Applications.
Privacy
Computational methods
-
8/6/2019 8-Pattern Recognition in Ubicom
11/29
Augmented Reality
The opposite approach from virtual reality. VR encloses people in an artificial world using
computers.
AR augments objects in the real world using
computers.
Examples Digitaldesk (Wellner 1993)
KARMA (Feiner 1993) Flatland (Mynatt 1999) augmented whiteboard
UbiTV, MRWindow, ARTable (Woo 2006)
-
8/6/2019 8-Pattern Recognition in Ubicom
12/29
Mobile AR
What is AR? To enhance the users
perception of and interactionwith the real world through
supplementing the real worldwith 3D virtual objects thatappear to coexist in thesame space as the realworld
We define AR system
Blends real and virtual, in areal environment
Interact on-the-fly Augment in 3D
-
8/6/2019 8-Pattern Recognition in Ubicom
13/29
Mobile AR ()ubiTV
A
Invisible
C
B
ubiTV
-
8/6/2019 8-Pattern Recognition in Ubicom
14/29
Mobile AR in U-Space
-
8/6/2019 8-Pattern Recognition in Ubicom
15/29
Future of Ubicomp
A key concept of ubicomp is to usetechnology to create a calmerenvironment (Weiser 1998).
Computer technology should servehumans as environment that does not
occupy much of human attention,and should serve humans calmly notconsuming human effort.
-
8/6/2019 8-Pattern Recognition in Ubicom
16/29
Pattern Recognition inUbicomp
Personalization Need to provide personalized service based on
each persons service history and preference.
Collection of user
s data using wireless sensors Recognition of users behavior pattern
Object or image recognition to get the
augmented data in U-space. Speech recognition
-
8/6/2019 8-Pattern Recognition in Ubicom
17/29
Process of PatternRecognition
sensorsensor Featuregeneration
Featureselection
Classifierdesign
Systemevaluation
pattern
-
8/6/2019 8-Pattern Recognition in Ubicom
18/29
Speech Recognition
Isolated word recognition (IWR) Continuous speech recognition (CSR)
Speaker-dependent recognition
Speaker-independent recognition
Dynamic time warping (DTW) - DP
finds an optimal match between two sequencesof feature vectors which allows for strechedand compressed sections of the sequence.
-
8/6/2019 8-Pattern Recognition in Ubicom
19/29
Symmetrical DTW
-
8/6/2019 8-Pattern Recognition in Ubicom
20/29
Dynamic Time Warping
Global constraints Local constraints
Monotonicity - matching paths cannot go
backwards in time End point constraints
Starts at (0,0) and ends at (,J) and whose first
transition is to the node (1,1) The cost for the transitions
Euclidean distance
-
8/6/2019 8-Pattern Recognition in Ubicom
21/29
DTW Global Constraint
-
8/6/2019 8-Pattern Recognition in Ubicom
22/29
DTW Local Constraints
-
8/6/2019 8-Pattern Recognition in Ubicom
23/29
DTW Local Constraints
Sakoe and Chiba local constraints
-
8/6/2019 8-Pattern Recognition in Ubicom
24/29
-
8/6/2019 8-Pattern Recognition in Ubicom
25/29
Test Words
-
8/6/2019 8-Pattern Recognition in Ubicom
26/29
Feature Generation
Reference and test patterns 0.45, 0.4 sec. long
22050 Hz sampling rate
Frame size 512 samples long
100 samples overlapped
24, 21, 23 (Number of frames)
-
8/6/2019 8-Pattern Recognition in Ubicom
27/29
Feature Selection
r(j), j=1,,J - reference pattern t(i), i=1,, - test pattern
xj(n), n=0,
511 -
samples for thejth frame of the reference pattern.
Taking DFT
Xi(m)=xi(n)exp(-j2mn/512)/5121/2,m=0,,511
Use the first 50 DFT coeff. as features
r(j)=[Xj(0) Xj(1) Xj(49)]T, j=1,J
-
8/6/2019 8-Pattern Recognition in Ubicom
28/29
Cost
Euclidean distance between r(j) andt(i) corresponding to node (i,j) d(i,j)||r(j)-t(i)||
Symmetrical DTW D(i,j) = min[D(i-1,j-1),D(i-1,j),D(i,j-1)]
+d(i,j)
Asymmetrical DTW D(i,j) = min[D(i-1,j-2),D(i-1,j-1),D(I-
1,j)]+d(i,j)
-
8/6/2019 8-Pattern Recognition in Ubicom
29/29
Results
For the test pattern love Dlove = 11.472 (0.221)
For the test pattern kiss
Dkiss = 25.155 (0.559)
Dlove < Dkiss - correct