Transcript
Page 1: Tracklet-global Motion Cost Model

• Tracklet-global Motion Cost Model

GMCP-Tracker: Global Multi-object Tracking Using Generalized Minimum Clique Graphs

Amir Roshan Zamir, Afshin Dehghan, Mubarak Shah ({aroshan| adehghan | shah @cs.ucf.edu} , University of Central Florida)

1. Problem: Tracking by Detection

Key Contributions: A new tool in Computer Vision:

Generalized Minimum Clique Problem (GMCP): Typically useful where there are multiple possibilities for some

subproblems, as well as a global criterion to satisfy. GMCP utilized for computing tracklets and trajectories.

A new temporally global approach to Data Association A new tracklet-global motion cost model. Shifting approximation from Time Domain to Object Domain:

Finding tracklets/trajectories in a temporally global way. Finding the tracklet/trajectory of one object at a time (greedy).

Bipartite Matching vs. GMCP

5. Merging Tracklets into Trajectories:

ECCV ‘12UCF

2. Block Diagram:

3. Generalized Minimum Clique Problem (GMCP):

𝐺=(𝑽 ,𝑬 ,𝑤) 𝐺𝑠=(𝑽 𝑠 ,𝑬 𝑠 ,𝑤𝑠)

Generalized Minimum CliqueInput to GMCP

Frame 1 Frame 2 Frame 3

Frame 4 Frame 5 Frame 6

Frame 1 Frame 2 Frame 3

Frame 4 Frame 5 Frame 6

Frame 2

Frame 5

Frame 1 Frame 2 Frame 3

Frame 4 Frame 5 Frame 6

Frame 1 Frame 2 Frame 3

Frame 4 Frame 5 Frame 6

4. Finding Tracklets Using GMCP:

Inpu

t Det

ectio

nsIn

put G

raph

GM

inim

um C

lique

O

cclu

sion

Han

dlin

g U

sing

HN • Occlusion Handling Using Hypothetical Nodes:

6. Experimental Results

• Found Tracklets in Different Segments:

Minimum Clique

Trajectories

Input Tracklets Occlusion Handling using HN

Frame 1 Frame 2 Frame 3

Frame 4 Frame 5 Frame 6

𝑣11𝑣2

1

𝑣31

𝑣12

𝑣22

𝑣32

𝑣13 𝑣2

3

𝑣33

𝑣14

𝑣24

𝑣34

𝑣15

𝑣25

𝑣35 𝑣1

6𝑣26

𝑣36

𝐶1❑ 𝐶2

❑ 𝐶3❑

𝐶4❑ 𝐶5

𝐶6❑

Definition: (nodes≡detections) (clusters≡frames) (GMCP solution=tracklet

of one Object)

𝑣𝟒6

Proj

ect P

age

YouT

ube

Project Page: http://vision.eecs.ucf.edu/ projects/GMCP-Tracker/

Top Related