Presentation is loading. Please wait.

Presentation is loading. Please wait.

Tracking Pattern Recognition

Similar presentations


Presentation on theme: "Tracking Pattern Recognition"— Presentation transcript:

1 Tracking Pattern Recognition
GSI, December 2nd, 2010 Brief summary of Tracking Pattern Recognition status in Panda Gianluigi Boca Gianluigi Boca, GSI, 2 Dec 2010

2 The detectors for which we have presently a Pattern Recognition
code are the central detectors : Mvd, Stt, Tpc, Gem and the Muon Drift Tubes (MDT) Still missing is the Forward Tracker (Straws) because software development began later than the other detectors (in spring of this year). Ideal Pattern Recognition : hits are associated to tracks according to MC truth Real Pattern Recognition : hits are associated to tracks as in real life situation, with PR algorithm. Gianluigi Boca, GSI, 2 Dec 2010

3 Ideal Pattern Recognition
There has been Ideal PR for all Central Detector since a long time (at least more than one year ago). Also existing : 1) Ideal PR for Mvd + Tpc, based on Rieman sphere algorithm, and it works fine. 2) LHeTrack general purpose pattern recognition package (works well in region where B field is actually constant ) Gianluigi Boca, GSI, 2 Dec 2010

4 Real Pattern Recognition
Two approaches in Panda : ‘homemade’ code : Panda developers wrote code based on various Pattern Recognitions techniques. usage of LHeTrack, a general purpose PR code written by Russian scientist not active at GSI any more. Gianluigi Boca, GSI, 2 Dec 2010

5 Real Pattern Recognition
Mvd alone Homemade code : PR based on Riemann sphere; track vertex doesn’t have to be in (0,0,0). Efficiency : see Tobias’s talk at this meeting for efficiencies etc. LHeTrack code : works well with Mvd detector alone; track vertex has to be near (0,0,0). Gianluigi Boca, GSI, 2 Dec 2010

6 Real Pattern Recognition
Stt alone Homemade : PR based on 2 step process; use of a conformal mapping, ‘road finding’ procedure, MILP software (GLPK library) for fast minimization algorithm AND for solving the straw left/right ambiguity. Vertex must be within  5 mm from (0,0,0). See Gianluigi’s talk at March 2010 collaboration meeting LHeTrack code : doesn’t work well with Stt detector alone; this is probably due to the fact that Stt hits are not XYZ hits. Gianluigi Boca, GSI, 2 Dec 2010

7 Real Pattern Recognition
Tpc alone Homemade : PR based on Rieman sphere ; vertex doesn’t have to be in (0,0,0). Also a version exists with Hough trasform method. See Felix’s talk at September 2009 meeting. LHeTrack code : works well with Tpc detector alone; track vertex has to be near (0,0,0). Gianluigi Boca, GSI, 2 Dec 2010

8 Real Pattern Recognition
Gem alone Homemade : match hits in indiviual stations; find pairs of hits in adjacent stations ( track segments) from possible tracks coming from vertex in (0,0,0) by using parabolic formula ( = fast to compute) ; merge track segments into a track; removal of spurious hits; remove obviously wrong tracks. See Radek’s talk at December 2009 meeting. LHeTrack : it doesn’t works well with Gem detector alone; this may be due to the non-uniformity of B field towards the edge of the solenoid. Gianluigi Boca, GSI, 2 Dec 2010

9 Real Pattern Recognition
Muon DT alone Homemade : PR based on a proximity search starting from inner hits outwards. See Stefano’s talk at December 2009 meeting. Pattern Recognition MdtHit from inner layer one tracklet PndMdtTrk closest hit in next layer in a search cone and so on… Enccap and Muon Filter threated as single module LHeTrack code : it doesn’t make sense to use it in this case

10 Real Pattern Recognition
Mvd + Tpc Homemade code : PR based on Riemann sphere Mvd code extended to include Tpc hits; track vertex doesn’t have to be in (0,0,0). Code becomes slow due to high combinatorics. LHeTrack code : works well also with Mvd + Tpc ; track vertex has to be near (0,0,0). This is the code practically used at present. Gianluigi Boca, GSI, 2 Dec 2010

11 Real Pattern Recognition
Mvd + Stt Homemade : PR Stt alone; tracks extrapolated in Mdv region and new hits associated; new fit (in 2 steps) ; use of a conformal mapping; MILP software (GLPK library) for fast minimization algorithm AND for solving the straw left/right ambiguity. Vertex NOT constraind to originate from (0,0,0). Much better rejection of spurious hits. Better resolution. see Gianluigi’s talk at September 2010 meeting. LHeTrack code : doesn’t work well with Mvd + Stt detector; this is probably due to the fact that Stt hits are not XYZ hits. Gianluigi Boca, GSI, 2 Dec 2010

12 Real Pattern Recognition
Mvd + Gem Homemade : PR code ready (see Radek’s talk at this meeting). Gianluigi Boca, GSI, 2 Dec 2010

13 Real Pattern Recognition
All central detectors LHeTrack code : in principle it works with any set of hits whose position is defined by X, Y, Z with small errors. It works with the assumption of a constant B field  helix trajectory. No wonder it tends to fail in non-uniform B field regions and for the Stt (hit X, Y, Z need to be calculated by prefit). Gianluigi Boca, GSI, 2 Dec 2010

14 Real Pattern Recognition, work in progress
Mvd + Stt + Gem in overlap region The overlap region between Stt and Gem goes from 10° <  < 20° in the ‘long’ Stt + 3 Gem geometry and 14° <  < 21° in the ‘short’ Stt + 4 Gem geometry. Lia is writing this code : starting from Mvd + Stt homemade PR it projects tracks onto Gem sensitive planes (with Geane) and looks for hits to associated to. With new ‘extended’ track it redoes the track fit. Gianluigi Boca, GSI, 2 Dec 2010

15 Status of Ideal Pattern Recognition & Track Fitting Software
Done for : In progress Mvd alone Stt alone Tpc alone Gem alone Forward Straws Mvd + Tpc Mvd + Tpc + Gem Muon DT alone Mvd + Stt + Gem Also : LHeTrack general purpose pattern recognition package (works well in region where B field is actually constant ) For track fitting : Genfit (general Kalman package) ; Stt alone Kalman code; Stt alone Helix fit Gianluigi Boca, GSI, 2 Dec 2010

16 Status of Real Pattern Recognition & Track Fitting Software
Existing Pattern Recognition software for : Detectors alone Combined Detectors Mvd alone Stt Mvd + Tpc with Riemann (but slow) Mvd + Tpc with LHeTrack Mvd + Gem (see Radek’s talk) Muon DT alone Mvd + Stt Also : LHeTrack general purpose pattern recognition package; works well in region where B field is constant but not for Stt Tpc alone Gem Work in progress : Stt + Gem in overlapping region of  For track fitting : Genfit (general Kalman package) ; Stt alone Kalman code; Stt alone Helix fit Gianluigi Boca, GSI, 2 Dec 2010


Download ppt "Tracking Pattern Recognition"

Similar presentations


Ads by Google