close

Вход

Забыли?

вход по аккаунту

?

Mushroom, fish and Classification machines With focus on linear

код для вставкиСкачать
Detection, tracking and sizing of
fish of in data from DIDSON
multibeam sonars
Helge Balk1, Torfinn Lindem1, Jan KubeДЌka2
1 Department of Physics, University of Oslo, PO.Box.1048. Blindern, NO-0317 Oslo,
Norway email: helge.balk@fys.uio.no, Torfinn.lindem.@fys.uio.no
2 Biology Centre of Czech Academy of Sciences, Institute of Hydrobiology, Na sadkach
7, CZ 37005 Ceske Budejovice, Czech Republic. e-mail: kubecka@hbu.cas.cz,
1
Introduction
Conclusion
Detection
methods
CFD AND
DIDSON
Echogram
approach
tracking
Inc.Video
methods
3D
approach
2
University
of Oslo No
Placing Norway on the map
Biological
institute Cz
3
Our main interest
пЃµ As
usual to find out abot the fish
пЃµ How
many
пЃµ How big
пЃµ What are they doing
4
Equipment that may be used
DIDSON
Simrad SM2000
Coda Octopus
Echoscope
Split beam
Resons-Seabat
Simrad MS70
DIDSON
пЃµ Dual
frequency Identification SONnar
пЃµ Developed for military underwater tasks like
diver night vision and mine searching
пЃµ Become
popular for fish studies
пЃµ Identification
ability
пЃµ Can see pictures of the fish.
пЃµ Fish size from geometry, not from TS
6
Our aim
пЃµ Develoop
пЃµ
a target detector for DIDSON data
Can vi use the Cross Filter Detector CFD
develooped for ordinary echogram
пЃµ If
not, can we optimise it to fit the DIDSON data
пЃµ Or
is there something to learn from the video
world
7
Dual-Frequency Identification
Sonar (DIDSON)
8
DIDSON problems
пЃµ Low
snr,
пЃµ Low
dynamic span,
пЃµ Not
calibrated,
пЃµ Not
veldefined sample volume
пЃµ Only
x,z, but no y position information
9
DIDSON inside
10
Examples of data
11
Introduction
Detection
methods
Conclusion
CFD AND
DIDSON
Tracking
Aim, material
and methods
3D approach
Echogram
approach
12
Detection theory - methods
пЃµ Edgebased
пЃµ Gradient
operators
пЃµ Linking Edge
пЃµ
Stastistical
пЃµ
Relaxation
пЃµ
If this is a fish pixel,
then…
пЃµ Thresholding
пЃµ Constant,
пЃµ Addaptive,
13
Cross Filter Detector (CFD)
Filter 1
Filter 2
Variance
a
b
c
Comparator
Evaluator
Traces
Filter
direction
Signal a
Signal b
Signal c
Combine
Evaluator
CFD –Addaptive thresholding
Main challenge: Find the optimal threshold
signal
threshold
Detection methods
How to fit the Crossfilter to video like data?
Can we learn something from the video world?
Echogram
Foreground filter
Crossfilter
detector
Background filter
Comparator
Evaluator
variance
Comparator
Video
Common video
processing
Evaluator
Background
Modelling
16
Background modelling.
– the most important part.
пЃµ
пЃµ
Non recursive
пЃµ
пЃµ
пЃµ
пЃµ
Previous picture
Median
Linear predictive
Nonparametric
пЃµ
Approximated median
пЃµ
Kalmann filter
пЃµ
Mixture of Gausians
Comparator
Video
Common video
processing
Recursive
Background
Modelling
Evaluator
Background modelling.
– the most important part.
Ching , Cheung and Kamath found
пЃµ
Three best
пЃµ
1 Mixture of Gausians
пЃµ
2 Median
пЃµ
3 Approximated median
пЃµ
пЃµ
Not much difference
App. Median much faster and
simpler than the others
Sen-Ching S. Cheung and Chandrika Kamath Center for Applied Scientic
Computing Lawrence Livermore National Laboratory, Livermore, CA 94550
18
Comparator
Comparator
Video
Common video
processing
Evaluator
Background
Modelling
19
Evaluator
пЃµ
пЃµ
пЃµ
пЃµ
Morfological filter
Recognise fish on size and shape
May use higher order statistics
Connect parts of targets
Comparator
Video
Common video
processing
Evaluator
Background
Modelling
20
Introduction
Summary
Detection
methods
CFD AND
DIDSON
Echogram
approach
Tracking
Inc.Video
methods
3D
approach
21
Echogram approach
Amplitude
Detector
Gain
96-Ch
Multi пѓ 1 beam
Echogram
generator
Multi beam-viewer
Amp-Echogram
22
Generate echograms and apply the
Cross-Filter
How to combine many beams into one ?
a) Mean echogram
пЃµ
At each range bin extract mean values from a selected number
of beams. Like an ordinary transducer with controllable
opening angle
b) Max Intensity
пЃµ
At each range bin, select the sample from the beam with
highest intensity
23
Data recorded by Debby Burwen
Generating Echograms from multi beam
Many beams пѓ a) Averaging a number of beams 10x12 deg
1 beam
b) Pick the beam with strongest intensity
24
Echogram approach
Testing the CFD on many to 1
beam echograms
25
Echogram approach
Echogram approach works well
until density becomes too high
We want to push the density limit
26
Introduction
The original
Cross filter
Summary
CFD AND
DIDSON
Tracking
Aim, material
and methods
3D approach
Echogram
approach
27
3D approach
Adding a third dimension
пЃµ Work
directly on the multi beam data
пЃµ Want
to detect more than one target in
the same range bin
range
range
time
width
time
2d-trace
3d-trace
28
3D approach
We added the beam dimension to
the filters
Beam. nr
Running window operators
Ping
2D
Range
3D
Ping
Range
New
DDF
29
Test foreground
filter
operator size
Frame 513
Beam
Range 513
513
Test Background
filter
operator size
Frame
Beam
Range
1
1
525
15
1
Testing cross filter on a small
trout in Fisha River
Max Intensity echogram
32
CFD with filters
and threshold
Forefilt 3 x 3 x 3
Back filt 3 x 3 x 3
Threshold Offset=20
Evaluator can take away
unwanted targets
34
35
Introduction
Summary
Detection
methods
CFD AND
DIDSON
Echogram
approach
Tracking
Inc.Video
methods
3D
approach
36
Extended the background filter
with an approximated median
operator
(N. McFarlane and C. Schoeld 1995)
If Q FBR пЂѕ AM
If Q FBR пЂј AM
BR
BR
then
AM
then
AM
BR
BR
пЂ«1
пЂ­1
Q
ddf
37
And extended the
comparator with
alternatives
Threshold
Foreground
a
Background
b
If ( a - b )>T )
detection
38
Background
subtraction
Forefilt 3 x 3 x 3
Back filt 3 x 3 x 3
App.Median
Threshold Offset=20
Introduction
Summary
Detection
methods
CFD AND
DIDSON
Echogram
approach
Tracking
Inc.Video
methods
3D
approach
40
The initial idea was to detect
traces directly by clustering
Cluster of overlapping
fish pictures
( Work well in the echogram approach )
41
But data often showed traces split
up in individual fish pictures
Center of gravity
Tracker needed
for fast fish
Clustering worked
for big slow fish
track
42
Special predictor can be made
for multi beam data
In addition to traditional predictors are
available such as Alpha Beta and Kalman
Special predictor can be formed
from the DIDSON fish picture
Fish center line
predictor
43
Introduction
Summary
Detection
methods
CFD AND
DIDSON
Echogram
approach
Tracking
Inc.Video
methods
3D
approach
44
Summary
Echogram
Foreground filter
Crossfilter
detector
Background filter
Comparator
Evaluator
variance
Comparator
Video
Common video
processing
DIDSON
Background
Modelling
3D-Foreground filter
Best method
Evaluator
Comparator
Evaluator
Tracker
Background
Modelling
45
Best method for moving targets
Summary
пЃµ
пЃµ
Needed in most cases
Need for various predictors
depending on data
( a - b )>T )
DIDSON
3D-Foreground filter
a
Comparator
Evaluator
Tracker
b
3D better than 2D
Optimise on improving
foreground
Background
Modelling
Improved
foreground signal
Approximated
Median
46
Run demo now if time
47
And that was it! Thanks for
the attention! Questions?
Introduction
Summary
Detection
methods
CFD AND
DIDSON
Echogram
approach
Tracking
Inc.Video
methods
3D
approach
48
Документ
Категория
Презентации
Просмотров
12
Размер файла
7 204 Кб
Теги
1/--страниц
Пожаловаться на содержимое документа