Tracking

1
Tracking
2
What is tracking?

• Finding the trajectories of objects over time
• Xi (t)
• Long military tradition
• Radar
• Different sensors
• Sampling rate
• Often position
• Trajectories in
• n-dim. space
• Active tracking versus passive tracking

3

• Predict, match, and update - framework
• Update Combine predicted and measured data (past
  and present)
• Prediction Ease the matching and limit search
  area

States
4

• Key questions
• Do I expect multiple objects?
• Can I assume a one-to-one mapping?
• Which similarity measure should I use?
• Should I weight the different features?
• What is the effect of the two error types (CTH)?
• New and lost objects
• Will merge/split (occlusion) situations occur?
• Should I use a local or a global matching
  strategy?
• Use the Relaxation method?
• The key thing to remember from this mini-module
• Tracking gt Predict, match, and update
  

5
What could we track?

• Points
• Objects
• Blobs or
• Specific objects (such as faces)
• Planes

6
First, Point Tracking

• (weve done this before, when?)

7
Textured area
8
Edge
9
Homogenous area
10
KLT tracking

• Select feature by
• Monitor features by measuring dissimilarity

11
Blob TrackingMean Shift 03, Comaniciu, Meer
12
Contour TrackingCONDENSATION 98-02, Isard,
Toyama, Blake
13
Today, Blob Tracking

• What is the state space of a blob?
• What is the feature space of a blob?

14
Mean-Shift Object TrackingGeneral Framework
Target Representation
15
Mean-Shift Object TrackingGeneral Framework
Target Localization
Start from the position of the model in the
current frame
Repeat the same process in the next pair of frames
16
Mean-Shift Object TrackingTarget Representation
Kernel Based Object Tracking, by Comaniniu,
Ramesh, Meer
17
Mean-Shift Object TrackingPDF Representation
Target Model (centered at 0)
Target Candidate (centered at y)
Q is the target histogram, P is the object
histogram (depends on location y)
18
Mean-Shift Object TrackingFinding the PDF of the
target model
Target pixel locations
Weighted by mean shift Kernel
19
Mean-Shift Object TrackingSimilarity Function
Target model
Target candidate
Similarity function
20
Mean-Shift Object TrackingTarget Localization
Algorithm
Start from the position of the model in the
current frame
21
Mean-Shift Object TrackingApproximating the
Similarity Function
Model location
Candidate location
Independent of y
Density estimate! (as a function of y)
22
Mean-Shift Object TrackingMaximizing the
Similarity Function
The mode of
sought maximum
23
Mean-Shift Object TrackingApplying Mean-Shift
The mode of
sought maximum
Original Mean-Shift
Find mode of
using
24
Mean-Shift Object TrackingAdaptive Scale
Problem
The scale of the target changes in time
The scale (h) of the kernel must be adapted
25
Mean-Shift Object TrackingResults
Feature space 16?16?16 quantized RGB Target
manually selected on 1st frame Average mean-shift
iterations 4
26
Mean-Shift Object TrackingResults
27

• What next?
• Points, blobs, and more complicated shapes have
  different
• State-spaces
• Update rules to go from one frame to the next
• What else comes into tracking?
• (matlab dirty tricks!)

28
Kalman Filter (1)
Welch, Bishop. An Introduction to the Kalman
Filter. SIGGRAPH01
http//www.cs.unc.edu/welch/kalman
29
Kalman Filter (2)

• Estimate a process
• with a measurement

x??n ... State of the process z??m ...
Measurement p, v ... Process and measurement
noise (zero mean) A ... n x n Matrix relates the
previous with the current time step B ... n x l
Matrix relates optional control input u to x H
... n x m Matrix relates state x to measurement z
30
Kalman Filter (6)
http//www.cs.unc.edu/welch/kalman