ROBUST VISUAL TRACKING A Brief Summary

Post on 12-Jan-2016

35 views 0 download

Tags:

description

ROBUST VISUAL TRACKING A Brief Summary. 1. Gagan Mirchandani School of Engineering, University of Vermont. 1. And Ben Schilling, Clark Vandam, Kevin Haupt. Algorithms from [1],[2]. Examples from [2]. Videos from [3]. [1] J.Wright, A.Y.Yang, A.Ganesh, S.S.Sastry and Y.Ma, - PowerPoint PPT Presentation

transcript

1

ROBUST VISUAL TRACKING

A Brief Summary

Gagan MirchandaniSchool of Engineering, University of Vermont

1

1 And Ben Schilling, Clark Vandam, Kevin Haupt

2

[1] J.Wright, A.Y.Yang, A.Ganesh, S.S.Sastry and Y.Ma, "Robust Face Recognition via Sparse Representation" IEEE Trans. PAMI , Feb. 2009, Vol.31, Issue:2, pp.210-227.

[2] X.Mei and H.Ling, "Robust Visual Tracking and Vehicle Classication via Sparse Representation" IEEE Trans. PAMI , Nov. 2011, Vol.33, Isssue:11, pp.2259-2272.

[3] Ben Schilling, Clark Vandam, Kevin Haupt

Algorithms from [1],[2]. Examples from [2].Videos from [3].

3

1. Introduction

• Background, Goals

• Tracking and Recognition - important topics in Computer Vision

• Studied for decades

4

2. Problem Areas

•• Tracking, recognition and counting objects (pedestrians, vehicles, bicyclists, etc. etc.)

• Needed for Policy determination, optimal traffic management, reduction of fuel, CO2 emission, etc.

• Needed for Surveillance

• Needed for Robotics

5

3. Challenges

• Occlusion, noise, cluttered real-world environment

• Illumination change, many people, changing pose

• Changing background, real-time online implementation

• Computational complexity grows exponentially

6

4. Theory

Basic problem: Given measurements y - Find x

7

Bayesian State Estimation

If f and h linear (and noise Gaussian) then we get the Kalman filter

8

Target candidate represented as sum of 10 templates (from previous frames) and trivial templates

9

10

Estimation Method•Particle filters numerically generate

the particles

• according to the pdf

•This is tracking. The particle filter propagates sample pdfs over time

•Computational effort often a bottleneck

11

5. Examples & Videos

12

Target candidate represented as sum of 10 templates (from previous frames) and trivial templates

13

Person walking; passing pole, high grass, body movement, occlusion.

14

Fast moving car with significant scale changes

Video taken from car in the back. Doll has pose & scale change and occlusion

16

L1, MS, CV, AAPF & ES Trackers

17

Drastic illumination change

18

Partial occlusion, background clutter

19

Severe occlusion

20

Face rotates 180 . Car moves out of frame.o

21

QuickTime™ and a decompressor

are needed to see this picture.

QuickTime™ and a decompressor

are needed to see this picture.

22

QuickTime™ and a decompressor

are needed to see this picture.

QuickTime™ and a decompressor

are needed to see this picture.

23

24

Questions?