t racking l earning d etection

36
Tracking Learning Detection A Camera that Learns.

Upload: fadhila

Post on 22-Feb-2016

31 views

Category:

Documents


0 download

DESCRIPTION

T racking L earning D etection. A Camera that Learns. Why is object tracking so important ?. vehicle detection. motion capture. r obot vision. scene analysis. hand recognition. surveillance. Outline. Why TLD?. - Traditional methods- Challenges to old ways. What is TLD intuitively?. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: T racking L earning D etection

TrackingLearningDetection

A Camera that Learns.

Page 2: T racking L earning D etection

Why is object tracking so important ?

surveillance

hand recognitionscene analysisve

hicl

e de

tecti

on

robot vision

motion capture

Page 3: T racking L earning D etection
Page 4: T racking L earning D etection

Outline

- Framework - Key ideasWhat is TLD intuitively?

How does TLD work exactly?

Experiment Results

Why TLD?- Traditional methods - Challenges to old ways

- Tracking - Learning - Detection - TLD

Page 5: T racking L earning D etection

Why TLD ?

Page 6: T racking L earning D etection

Why TLD ?

Page 7: T racking L earning D etection

Why TLD ?- Traditional method 1 :Tracking

Accumulate error.Fail if the object disappears.

Only need initialization.Appearance changing & Partial occlusions. (?)

Estimate motion over frames

Page 8: T racking L earning D etection

Why TLD ?- Traditional method 2 :

DetectionOff-line

Learning

Do not fail if the object disappears.

Need off-line training.Appearance changing & Partial occlusions.

Scan every frame independently

Page 9: T racking L earning D etection

Why TLD ?- Challenges to old ways• Not fail if the object disappears.

• Appearance changing.

• Partial occlusions.

Page 10: T racking L earning D etection

Why TLD ?- Challenges to old ways• Not fail if the object disappears.

• Appearance changing.

• Partial occlusions.

TLD is an algorithm

that can solve all these tasks !

Page 11: T racking L earning D etection

What is TLD intuitively?

Page 12: T racking L earning D etection

What is TLD intuitively?- Framework

Tracking

DetectionOff-line

Learning

Page 13: T racking L earning D etection

What is TLD intuitively?- Framework

Tracking Detection

On-lineLearning

Update

LocationTrajectoryfragment

Update

Page 14: T racking L earning D etection

What is TLD intuitively?- Key ideas 1 : Learning

Track & detect

Track

Detect

Updatestraining examples

(add/delete)

Update detector

•Appearance changing.

Page 15: T racking L earning D etection

What is TLD intuitively?- Key ideas 2 : Updating

True trajectory Tracking Detection

TLD

• Not fail if no object.

• Partial occlusions.

Page 16: T racking L earning D etection

How does TLD work exactly?

Page 17: T racking L earning D etection

How does TLD work exactly?- Framework

Page 18: T racking L earning D etection

How does TLD work exactly?- Tracking

1.Reliable point selection:

Choose point 1

Discard point 2

- Use forward-backward selection to select 50% of the most reliable points.

- Forward-backward selection

Page 19: T racking L earning D etection

How does TLD work exactly?- Tracking

2.Median-Flow tracker:

Tracks the reliable points between consecutive frames ,

and estimates displacement using median with confidence proportional to (1/scale of the median).

Page 20: T racking L earning D etection

How does TLD work exactly?- Framework

Page 21: T racking L earning D etection

How does TLD work exactly?- Detection

2bitBPfeatures

Scanningwindow

2bitBP features

Illuminationinvariance

Page 22: T racking L earning D etection

How does TLD work exactly?- Detection

RandomForest

Posteriors

2bitBPfeatures

Scanningwindow

Random Forest

If the average posteriors is larger than 50%, consider it an object

Page 23: T racking L earning D etection

How does TLD work exactly?- Detection

1-NN classifierConfidence =d-/(d-+d+)Random

ForestPosteriors

2bitBPfeatures

Scanningwindow

Feature space

Page 24: T racking L earning D etection

How does TLD work exactly?- Framework

Pick the one with higher confidence score

Page 25: T racking L earning D etection

How does TLD work exactly?- Learning

Feature space

Page 26: T racking L earning D etection

How does TLD work exactly?- Learning

Track & detect

Track

Detect

Updatestraining examples

(add/delete)

Update detector

Page 27: T racking L earning D etection

How does TLD work exactly?- Learning :

1

6

10

Appearance changingproblem solved!

9

Page 28: T racking L earning D etection

How does TLD work exactly? - Recap : t=0

Page 29: T racking L earning D etection

How does TLD work exactly? - Recap : t=1

Decide by confidence

score

Page 30: T racking L earning D etection

How does TLD work exactly? - Recap : t=1

Page 31: T racking L earning D etection

How does TLD work exactly? - Recap : t=2

Remove if false detect

Page 32: T racking L earning D etection

Experiment Result

Page 33: T racking L earning D etection

Experiment Result

Page 34: T racking L earning D etection

Experiment Result

Page 35: T racking L earning D etection

Experiment Result

Page 36: T racking L earning D etection

Thank you