Tracking-Learning-Detection(TLD) Zdenek Kalal, Krystian Mikolajczyk, - - PowerPoint PPT Presentation

β–Ά
tracking learning detection tld
SMART_READER_LITE
LIVE PREVIEW

Tracking-Learning-Detection(TLD) Zdenek Kalal, Krystian Mikolajczyk, - - PowerPoint PPT Presentation

Oct 10, 2023 161 likes β€’480 views

Tracking-Learning-Detection(TLD) Zdenek Kalal, Krystian Mikolajczyk, Jiri Matas PAMI 2010 Presented by: Lyne P. Tchapmi Stanford University/CS231B Overview Problem Definition Previous Works Hinterstossier et al. Babenko et al.

slide-1
SLIDE 1

Tracking-Learning-Detection(TLD)

Presented by: Lyne P. Tchapmi Stanford University/CS231B

Zdenek Kalal, Krystian Mikolajczyk, Jiri Matas PAMI 2010

slide-2
SLIDE 2

Overview

  • Problem Definition
  • Previous Works
  • Hinterstossier et al.
  • Babenko et al.
  • Contribution
  • TLD Framework
  • Tracker
  • Detector
  • Learning
  • Performance Analysis
slide-3
SLIDE 3

Problem Definition

TRACKING

Bounding Box Video Frame Object Location

slide-4
SLIDE 4

Overview

  • Problem Definition
  • Previous Works
  • Hinterstossier et al.
  • Babenko et al.
  • Contribution
  • TLD Framework
  • Tracker
  • Detector
  • Learning
  • Performance Analysis
slide-5
SLIDE 5

Hinterstossier et al. CVPR 2009

Sparse Bayesian Learning for Efficient Visual Tracking

RVM: Relevance Vector Machine (modified SVM)

slide-6
SLIDE 6

Grabner et al. ECCV 2008

slide-7
SLIDE 7

Overview

  • Problem Definition
  • Previous Works
  • Hinterstossier et al.
  • Babenko et al.
  • Contribution
  • TLD Framework
  • Tracker
  • Detector
  • Learning
  • Performance Analysis
slide-8
SLIDE 8

TLD Contribution

  • New Tracking Framework (Tracking-Learning-Detection

TLD)

  • P-N Learning
  • Handles unknown objects
  • Long-term tracking
  • Detector resets tracker (avoids drift)
  • Detector improves over time
slide-9
SLIDE 9

Overview

  • Problem Definition
  • Previous Works
  • Hinterstossier et al.
  • Babenko et al.
  • Contribution
  • TLD Framework
  • Tracker
  • Detector
  • Learning
  • Performance Analysis
slide-10
SLIDE 10

Implementation: TLD in detail

slide-11
SLIDE 11

Implementation: Tracker

  • Pyramidal Lucas-Kanade Tracker (KLT)
slide-12
SLIDE 12

Implementation: TLD in details

slide-13
SLIDE 13

Implementation: Object detector

  • Scanning-window grid + Cascaded classifier

2-bit binary patterns Randomized Fern Forest

slide-14
SLIDE 14

Implementation: TLD in details

slide-15
SLIDE 15

Implementation: TLD in details

slide-16
SLIDE 16

Implementation: Memory

π‘ž+: π‘π‘π‘˜π‘“π‘‘π‘’ π‘žβˆ’: π‘π‘π‘‘π‘™π‘•π‘ π‘π‘£π‘œπ‘’

slide-17
SLIDE 17

Implementation: P-N Learning

slide-18
SLIDE 18

P-N-LEARNING

slide-19
SLIDE 19

Implementation: P-Expert

I’ve seen this before…. …add to model!

slide-20
SLIDE 20

Implementation: P-Expert

Add points on valid trajectory(solid) Update Core Initial Model

slide-21
SLIDE 21

Implementation: N-Expert

If it isn’t positive, it must be negative… …add to model!

slide-22
SLIDE 22

Implementation: TLD in details

slide-23
SLIDE 23

Overview

  • Problem Definition
  • Previous Works
  • Hinterstossier et al.
  • Babenko et al.
  • Contribution
  • TLD Framework
  • Tracker
  • Detector
  • Learning
  • Performance Analysis
slide-24
SLIDE 24

Performance Metrics

  • Precision

𝑄 = #𝑒𝑠𝑣𝑓 π‘žπ‘π‘‘ #π‘ π‘“π‘‘π‘žπ‘π‘œπ‘‘π‘“π‘‘

  • Recall

𝑆 = #𝑒𝑠𝑣𝑓 π‘žπ‘π‘‘ #π‘π‘‘π‘‘π‘£π‘ π‘ π‘“π‘œπ‘‘π‘“π‘‘ 𝑒𝑝 π‘šπ‘π‘π‘“π‘š

  • F-measure

𝐺 = 2𝑄𝑆 𝑄 + 𝑆

slide-25
SLIDE 25

Performance: CoGD Dataset

  • TLD achieves maximal possible score in all sequences
slide-26
SLIDE 26

Performance: Prost Dataset

  • TLD scores best in 9/10 sequences, outperforms 2nd best by 12%
slide-27
SLIDE 27

Performance: TLD Dataset

slide-28
SLIDE 28

Performance: TLD Dataset

  • TLD scores best on average 81%, vs 22% for 2nd best (F-measure)
slide-29
SLIDE 29

Limitations

  • Articulated Objects (pedestrians)
  • Full out of plane rotations
  • Tracker gets lost
  • Detector sees an appearance never seen in model
  • Tracker is fixed
  • Makes the same errors
slide-30
SLIDE 30

QUESTIONS?