1 stereographic analysis of coronal features for the stereo mission eric de jong, paulett liewer,...
DESCRIPTION
3 Tiepointing Tools for Triangulation of Solar Features Tiepointing by Hand & Eye Use commercial software (ENVI) on conventional workstation Use 3D Cursor Tiepointing Tool (developed at JPL) –Needs workstation supporting stereo viewing Tools tested using synthetic stereo image pairs Tiepointing tools to locate the same “feature” in both images Present Research: Automatic Tiepointing using Automatic Feature Tracking Feature tracking for loops - test loops and real data Feature tracking for CMEs - real dataTRANSCRIPT
1
Stereographic Analysis of Coronal Features for the STEREO Mission
Eric De Jong, Paulett Liewer, Jeff Hall, Jean Lorre, Shigeru Suzukiand the SECCHI Team
STEREO Science Working Group,Berkley California
Outline • STEREO analysis using triangulation• Progress on Automatic Feature Tracking
•Test stereo loops and their 3D reconstruction•Coronal EUV loop feature tracking•Coronal Mass Ejection (CME) tracking
2
Determination of 3D Structure via Triangulation
Determination of 3D Geometry from Stereo Image Pairs
Triangulation: Determine 3D location of a point seen from two known locations
x= y’-ycosαsinα
α
x
z
y
x’
y’
x=x’cosα+y’sinα
y=y’cosα-x’sinαz=z’
Coordinαtes of two views relαted by simple rotαtionαl trαnsform
Stereo Imαges give y,y’ , Solve for x,x’
Coronαl loop viewed from two αngles sepαrαted by α
• In principle, two views determines com pletely (x,y,x) solαr coordinαtes of loop
• For sαm e “feαture” in tim e sequence of imαges, determine (vx,vy,vz)
• Technique limited by αbility to locαte sαm e “feαture” in both imαge
2
3
Tiepointing Tools for Triangulation of Tiepointing Tools for Triangulation of Solar FeaturesSolar Features
Tiepointing by Hand & Eye • Use commercial software (ENVI) on conventional
workstation• Use 3D Cursor Tiepointing Tool (developed at JPL)
– Needs workstation supporting stereo viewing• Tools tested using synthetic stereo image pairs
Tiepointing tools to locate the same “feature” in both images
Present Research: Automatic Tiepointing using Automatic Feature Tracking
•Feature tracking for loops - test loops and real data•Feature tracking for CMEs - real data
4
xyzsun
XYZSUN - 3D Solar Coordinates from Image TiepointsComputes transformation from solar coordinates to telescope coordinates
& projection on image planeUses software developed at JPL for planetary image processing
P’=M P + Rs/c
where M is transformation matrixP’ is point P in camera frame
Camera/Spacecraft Coordinate System
Solar Coordinate System
P
x’
z’
xy
zRs/cy’
Image plane at z’=-f
“Location” is point of closest approach of rays computed from 2 imagesS/C 1left eye image
S/C 2right eye image
ray 1
ray 2
Only perfect data would have two tiepoints map to exact same 3D location4
5
Sources of Error in Triangulation
Sources of Error in Triangulation
1. Ability to identify a feature in both images
Feature will look different from different angleIntegrated line-of-sight effects contribute to this
“Feature” may not be real – may be line-of-sight effect
2. Error in 3D determination depends on x-y error and angleResulting error in feature height z is magnified by 1/sinα (α=stereo αngle)
=> Error in height Δz/R sun = Δx/(R sunsin α)
Tαke error Δx = 1 pixel (requ ires excellent reg istrαtion αnd feαture iden tificαtion)
For STEREO/EUVI imαge with R sun ~ 700 pixels, Δx/R sun = 0.15%α= 15° => Δz /R sun ~ 0.6% Δz = 4200 kmα= 45° => Δz /R sun ~ 0.2% Δz = 1400 km
x-y (Δx) error is very sens itive to both knowledge o f spαcecrαftpointing αnd resolution of im αge
• Im plicαtion of 1+2 toge therAngles 15° < α < 20° m αy prove better thαn 25° < α < 35°
7
6
Automatic Feature Tracking for Coronal Loops Automatic Feature Tracking for Coronal Loops as seen in EUV and Soft X-rayas seen in EUV and Soft X-ray
A. Test Loops - Case 1 of 2 A. Test Loops - Case 1 of 2
SUN: 16May1994 CML=140°SUN: 17May1994 CML=125 °* Pattern on sphere shows magnetic field at photosphere (magnetogram)
Original Stereo Data - 2 Views of 3D Coronal Loops (angle=15º)Loops computed from measured solar magnetic fields at photosphere*
7
Results for Automatic Stereo Feature TiepointingResults for Automatic Stereo Feature Tiepointing
Algorithm: Follow along bright features Dark Segments on Loops are Matched Stereo Points*
SUN: CML=125 ° SUN: CML=140 °* Matched Stereo Points: Rays from the two points cross near the Sun
8
Comparison of Reconstructed 3D Loops with OriginalComparison of Reconstructed 3D Loops with Original3D loops reconstructured from tiepoints shown as colored
loops overlying original black loops
Successful Automatic 3D reconstruction of loops from stereo pairusing automatic tiepointing and XYZSUN
9
Automatic Feature Tracking for Coronal Loops Automatic Feature Tracking for Coronal Loops as seen in EUV and Soft X-rayas seen in EUV and Soft X-ray
A. Test Loops - Case 2A. Test Loops - Case 2
* Pattern on sphere shows magnetic field at photosphere (magnetogram)
Original Stereo Data - 2 Views of 3D Coronal Loops (angle=26º)Loops computed from measured solar magnetic fields at photosphere*
SUN: 3January1994 CML=96°SUN: 5January1994 CML=70 °
10
Results for Automatic Stereo Feature TiepointingResults for Automatic Stereo Feature TiepointingAlgorithm: Follow along bright features Dark Segments on
Loops are Matched Stereo Points*
* Matched Stereo Points: Rays from the two points cross near the Sun
11
Comparison of Reconstructed 3D Loops with OriginalComparison of Reconstructed 3D Loops with Original
Too many loops leads to false tiepointsRays from different loops in the two images happen to cross near SUN
3D loops reconstructured from tiepoints shown as colored loops overlying original black loops
12
Progress in automatic feature tracking
Progress in Automatic Feature TrackingGoal: Automatic location of “features” in two or more images
and creation of tiepoints for triangulation
Now developing using concepts of direction and Directionality:What direction of motion in image minimizes changes in intensity Ii?
7
q
I i
I i+1
Moving window centered αt pixel (x,y)
dθ = Ii +1 − Iiwindow
∑
θ − direction
dθ
π0
dmax
dmin
direc tion for each p ixel: θ of minimum d θ
Direct ional ity = D(x,y) = dmax-dmin812
13
Automatic Feature Tracking using DirectionalityFor each pixel have
• direction q which minim ize s chαnge in intensity regαrdless of intensity
• Δirectionαlity Δ(x,y) – me αsures how m uch this direction is preferre d
First stαge - find “feαtures” or “segm ents” in αn im αge:
1. Creαte αn im αge with Δirectionαlity Δ(x,y) αs the intensity of pixel (x,y)
2. Loop through pixels stαrting with highest Δ irectionαlity Δ
Creαte α feαture or “segm en t” by connecting to ne ighboring pixels(pixel window) with neαrly the sαm e dire ction q
Continue unless intersect αnother “segm en t”
End product: File of pixe ls for eαch segm ent/feαture9
14
Automatic Feature Tracking using DirectionalitySecond Stage – Finding same segment in second image
For each segment/feature in 1st image,Loop over pixels in the segment:
Create “bar code” intensity pattern for each pixel by movingperpendicular to direction q −− ”bαr code” is now α correlαtion“window”
Locαte pixel in 2nd im αge correcting for solαr rotαtion (SC motion)
Se αrch αround this pixel for α pixel with α “bαr code” with α highcorre lαtion
If correlαtion exceeds threshold, m αrk this pixe l αs sαm esegm en t/feαture
End product: file of pixe ls for correspondingsegm ent/feαture in 2nd im αge
10
15
Original TRACE Image Pair - 1 hour separation
trb_20001109_021021 trb_20001109_030008
16
Directionality Images
17
Traced Segments
18
Left: Traced Segments (Image 1) Right: Correlated Segments (Image 2)
19
Conclusions
• Stereoscopy (α.k.α. triαngulαtion) cαn be used todeterm ine 3Δ geom etry/locαtion of coronαl feαtures
• Tools αnd softwαre to determ ine 3Δ locαtion tested onsynthetic white light αnd EUV im αge pαirs
• Tie pointing “by hαnd” dem onstrαted using comm erciαlsoftwαre on conventionαl workstαtion s αnd in 3Δ usingSG I with stereo viewing using liquid crystαl goggles
• Δem onstrαted Automα tic Feαture trαcking betwee n twoimα ges using new m ethod bαsed on “Δ irection αlity”
Successful tests on TRACE dαtα with 15m in & 1hr sepαrαtions
Test with 2 hr sepαrαtions iden tified very few com m on feαtures
11
20
Store Organize Archive and real-time Retrieve (SOAR)
1000 [TB]
64 bit SunPeta Byte
Server(PBS) 1000 [TB]
PCSERVER
PC/MACAnalysis
Workstations
10 [TB]
100 [TB] 100 [TB]
PCSystem
10 [TB]
CPU-1
“““““
WirelessLaptops
1 [TB]
••
STORE
64 bit SunPeta Byte
Server(PBS)
RETRIEVE
CPU-1000
CPU-2 HDTVVIDEO
SERVER
ORGANIZE
ARCHIVE
100 [TB]
BeowulfSuperClusters(BCS)
CAST
21
Store Organize Archive and real-time Retrieve (SOAR) Visualization and Analysis Testbed VAT
FY % NodesHDTV [TB]DISK [TB]03040506070809101112
1 12 2
4060
482010
80100 10080080 8001000 1000
10 1020 2040 4 4080 8 8010100 10020020 20040040 40060060 600
22Zareh Gorjian combines a 3D terrain model, constructed from FIDO “Pancam” field test images, with Dan Maas’ rover model to create this simulated view of MER desert operations.
One work year was required to produce this 36 x Cluster Computing Mosaic Generation speed-up. A significant “Virtual Rover” challenge is to automate the speed-up and control of similar processes to achieve real-time performance.
Zooming in on the MER instruments enables us to see the level of detail available in a single photorealistic digital high resolution image.
Adapting Virtual Rover (VR) tools to SECCHI and STERO Instruments