Pattern Recognition in OPERA Tracking

1
Pattern Recognition in OPERA Tracking

• A.Chukanov, S.Dmitrievsky, Yu.Gornushkin

JINR, Dubna
OPERA collaboration meeting, Ankara, Turkey, 1-4
of April 2009
2
Outline

• Problems of pattern recognition in OPERA tracking
• Hough Transform method for straight track
  recognition
• Simple Tracing method for curved track
  reconstruction
• Spanning Tree method for curved tracks
• Status of proposed pattern recognition package.

3

• There is a problem in standard OpRelease pattern
  recognition algorithm.
• Mushower doesnt solve the problem but just
  serves as a patch for tracking algorithm.

Mushower extrap
Standard track
Pictures from Darios report 29/10/2008
4
Antoines presentation at LNGS end of 2008
5
A few examples from the latest RECO file
6
(No Transcript)
7
(No Transcript)
8
(No Transcript)
9
The OpRelease pattern recognition definitely
needs to be improved. The efficiente pattern
recognition method widely used in HEP experiments
(e.g. MINOS, ALICE, CBM, etc) is Hough transform
(HT) algorithm The algorithm is a part of the
BrickFinder and is fully integrated in the
OpRelease
10
Hough Transform for Straight Track Recognition
Hough transform uses representation of a line
in normal form This equation specifies a line
passing through point . That line is
perpendicular to the line drawn from the origin
to point in polar space. It can be shown
that in case of points belonging to the same line
and are constants.
For each of given point iteration
through different angles gives us
corresponding values of . Points are
saved in a 2D histogram. If there are some
straight tracks (or parts of tracks) in an event
there should exist distinct pikes in the
histogram. By determining of centers of gravity
of that pikes it is possible to
reconstruct parameters of track lines
by the following formulas
11
Example of HT Track Recognition event 234948251
OpRelease tracking solid line - Kalman
extrapolation, dash line - Mushower extrapolation.
Proposed HT track finding
Found
and
give track parameters
12
Example of HT Track Recognition event 234643825
OpRelease tracking solid line - Kalman
extrapolation, dash line - Mushower extrapolation.
Proposed HT track finding
Found
and
give track parameters
13
Example of HT Track Recognition event 234655944
OpRelease tracking solid line - Kalman
extrapolation, dash line - Mushower extrapolation.
Proposed HT track finding
Found
and
give track parameters
14
Example of HT Track Recognition event 234862308
OpRelease tracking solid line - Kalman
extrapolation, dash line - Mushower extrapolation.
Proposed HT track finding
Found
and
give track parameters
15
Example of HT Track Recognition event 234917207
OpRelease tracking solid line - Kalman
extrapolation, dash line - Mushower extrapolation.
Proposed HT track finding
Found
and
give track parameters
16
As shown in the given examples the muon track
is unambiguously distinguished by a pike in HT
histogram in case of so called difficult events.
Moreover, the result of Hough transform coincides
with Mushower extrapolation already at the
pattern recognition level (without a
fit). With 300 events with a muon found in
the CS (from Giovanni) a general performance was
estimated
17
(No Transcript)
18
(No Transcript)
19
(No Transcript)
20
(No Transcript)
21

• There are still events with muon track not found
  in the CS (out of list of Giovanni) (badly
  reconstructed in std tracking procedure?)

22
(No Transcript)
23
Houph Transform reconstruction of the same event
24
(No Transcript)
25
Houph Transform reconstruction of the same event
26
But sometimes due to low momentum of the particle
it is really difficult
27
Simple Tracing Method for Curved Track Finding
After the initial straight part of a track is
determined by a Hough transform in the beginning
of an event it is possible to find the rest tail
part of the track with help of proposed tracing
method (which in fact is a simplified kind of a
Kalman filter) 1) Finding a search direction
Linear fit on 7 last found hits of a track
2) Setting of search angle range Its own
angle range for each detector is used taking into
account its geometry and uncertainties. 3)
Finding hits in the following detector planes
inside the search angle range
Inefficiency of detectors (3 empty TT planes, 11
empty RPC planes) is taken into account. If
there are more than 1 candidates to track hits
only the hit accepted that is the nearest to
the search direction. 4) Including found hit to
the TrackElement and iterating steps 1-3 for
next planes or stop procedure in case of no hits
found
28
Backward Tracing with Background
When particles momentum is small the track can
be curved already in its beginning part. The
curved tracks are difficult for HT reconstruction
and even the simple tracing method can fail
within the shower environment. On the picture
below such a specific case is shown.
Event 217982179
Y, cm
T T planes
Line found by a Hough transform
track hits
wrong hits
There are no more sequential hits in the search
area
Z, cm
29
Example of Forward Tracing Procedure
Simple tracing along the beam direction works
easily (as shown on the picture) because there
are no background hits far away of the vertex.
Event 23356121
Y, cm
TT1
TT2
RPC2
RPC1
Z, cm
30
Spanning Tree Tracing Method for Track Selection
To solve such a problem it is useful to iterate
on all possible chains of the hits to select
among them the best chain. It can be done with
help of method of spanning tree tracing. It finds
different reliable track trajectories and than
consider the longest and most smooth chain of
hits to be the best track candidate.
Event 217982179
Y, cm
T T planes
As a result the longest and most smooth
track will be selected
Z, cm
31
Status of Proposed Pattern Recognition Package

• Event cleaning (removing of CT and isolated
  hits)
• done
• 2) Method of Hough Transform to find straight
  part of a track
• done
• 3) Tracing method to find curved tail part of a
  track
• done
• 4) Method of Spanning Tree Tracing to select the
  best track candidate within a shower
• done