Multiple Object Tracking Using Local PCA C. Beleznai 1 , B. Frhstck 2 - - PowerPoint PPT Presentation

▶
multiple object tracking using local pca
SMART_READER_LITE
LIVE PREVIEW

Multiple Object Tracking Using Local PCA C. Beleznai 1 , B. Frhstck 2 - - PowerPoint PPT Presentation

Oct 07, 2022 1.49k likes •1.73k views

Multiple Object Tracking Using Local PCA C. Beleznai 1 , B. Frhstck 2 , H. Bischof 3 1 Advanced Computer Vision GmbH ACV, Vienna, Austria 2 Siemens AG Austria, Program and System Engineering, Graz, Austria 3 Inst. for Computer Graphics and

slide-1
SLIDE 1
  • C. Beleznai1, B. Frühstück2, H. Bischof3

1Advanced Computer Vision GmbH – ACV, Vienna, Austria 2Siemens AG Austria, Program and System Engineering, Graz, Austria

  • 3Inst. for Computer Graphics and Vision, Graz University of Technology, Austria

Multiple Object Tracking Using Local PCA

slide-2
SLIDE 2

2

Motivation Motivation

Task: Task: tracking multiple objects in the space of observations tracking multiple objects in the space of observations Desired output: Desired output: consistent motion path consistent motion path

Complexity of tracking task:

missing observations interacting objects clutter, noisy observations

slide-3
SLIDE 3

3

Contents Contents

  • Introduction, related research

Introduction, related research

  • Input data

Input data

  • Trajectory segments by local PCA

Trajectory segments by local PCA

  • Trajectory segment linking

Trajectory segment linking

  • Results and evaluation

Results and evaluation

  • Conclusion

Conclusion

slide-4
SLIDE 4

4

Related Related research research

  • JPDAF Tracker (Bar

JPDAF Tracker (Bar-

  • Shalom1987)

Shalom1987)

  • Single stage approximation using fixed number of targets
  • Can not recover from failure
  • Multiple Hypothesis Tracker (Reid1979)

Multiple Hypothesis Tracker (Reid1979)

  • Heuristics to overcome computational complexity: pruning,

gating, N-scan back, k-best hypotheses

  • Monte Carlo

Monte Carlo methods methods ( (Vermaak Vermaak et. al 2003)

  • et. al 2003)
  • Promising performance in challenging scenarios

Data association following “deferred logic”

slide-5
SLIDE 5

Input data: Motion Input data: Motion-

  • based human detection

based human detection

  • Fast clustering of the difference image

Fast clustering of the difference image

  • Mean Shift procedure using integral images.
  • Model-based validation of hypothesized configurations:
  • Removing spurious detections
  • Occlusion handling
  • C. Beleznai, B. Frühstück and H. Bischof,

“Tracking Multiple Humans using Fast Mean Shift Mode Seeking”, PETS 2005 Workshop

slide-6
SLIDE 6

6

Related Related background background / / research research

Detecting objects (humans) by difference image clustering (C. Beleznai et al., ICIP 2004) Spatio-temporal data points (observations)

Prior information:

  • object size model H(x)
slide-7
SLIDE 7

7

Idea Idea

  • Motion of real

Motion of real-

  • world objects is subjected to kinematic

world objects is subjected to kinematic constraints constraints

  • Consequence: Observations at consecutive time

Consequence: Observations at consecutive time instances are strongly correlated. instances are strongly correlated.

LPCA LPCA

slide-8
SLIDE 8

8

Tracking Tracking by local by local PCA PCA

x t

(1) Selecting an initial point (2) Mean shift iterations to nearby mode (3) LPCA within the analysis window (5) Computing local anisotropy measure (4) Repeating from Step (2) (6) Stopping if:

  • no more data available
  • distribution shows no

anisotropy approaches 1 value of

slide-9
SLIDE 9

9

Motion model Motion model

  • The first eigenvector u1 can be interpreted as a velocity estimate
  • Simple update:
  • Computing data weights inversely proportional to

the distance between data and motion estimate

u1

  • Applying weighted local PCA
slide-10
SLIDE 10

10

Trajectory segment linking Trajectory segment linking

K generated trajectory segments

Greedy strategy to incrementally link segments (stopping criterion)

l(L) – length of a link S(L) – sum of angles: (α+β) / π

Constraints:

temporal ordering spatio-temporal smoothness

slide-11
SLIDE 11

11

Results Results – – Sequence Sequence 1 1

slide-12
SLIDE 12

12

Results Results – – Sequence Sequence 1 1

slide-13
SLIDE 13

13

Results Results – – Sequence Sequence 2 2

slide-14
SLIDE 14

14

Results Results – – Sequence Sequence 2 2 – – frame frame-

  • to

to-

  • frame

frame tracking tracking

slide-15
SLIDE 15

15

Results Results – – Evaluation of tracking performance Evaluation of tracking performance

Comparing to 42 annotated trajectories in 1013 frames: Comparing to 47 annotated trajectories in 2200 frames:

  • Norm. spatial deviation between

ground truth and measurement:

slide-16
SLIDE 16

16

Conclusions Conclusions

  • A simple and novel tracking approach.

A simple and novel tracking approach.

  • Two passes:

Two passes:

(1) LPCA (1) LPCA-

  • based trajectory segment generation,

based trajectory segment generation, (2) Trajectory segment linking. (2) Trajectory segment linking.

  • Tracker produces stable results at a low computational

Tracker produces stable results at a low computational demand. demand.

  • Possible improvements:

Possible improvements:

(1) combining forward and backward tracking, (1) combining forward and backward tracking, (2) hierarchical grouping of local trajectory estimates, (2) hierarchical grouping of local trajectory estimates, (3) (3) embedding embedding complementary complementary tracking tracking mechanisms mechanisms