Support Vector Machine based Logo Detection in Broadcast Soccer Videos Hossam M. Zawbaa Cairo University, Faculty of Computers and Information; ABO Research.

Support Vector Machine based Logo Detection in Broadcast Soccer Videos

Hossam M. ZawbaaCairo University, Faculty of Cairo University, Faculty of

Computers and Information; Computers and Information; ABO Research Laboratory; ABO Research Laboratory;

Cairo, EgyptCairo, Egypt

e-mail: [email protected]

Nashwa El-BendaryArab Academy for Science, Arab Academy for Science,

Technology, and Maritime Technology, and Maritime Transport; ABO Research Transport; ABO Research Laboratory; Cairo, EgyptLaboratory; Cairo, Egypt

e-mail: nashwa [email protected]

Aboul Ella HassanienCairo University, Faculty of Cairo University, Faculty of

Computers and Information; Computers and Information; ABO Research Laboratory; ABO Research Laboratory;

Cairo, EgyptCairo, Egypt

e-mail: [email protected]

Sang-Soo YeoDivision of Computer Division of Computer

Engineering, Mokwon Engineering, Mokwon University; KoreaUniversity; Korea

e-mail: [email protected]

m

Gerald SchaeferDepartment of Computer Department of Computer

Science, Loughborough Science, Loughborough University; U.K.University; U.K.

e-mail: [email protected]

g

By

Agenda

1- Introduction2- The Proposed System3- Dominant Color4- Shot-boundary detection5- Shot-type Classification6- Replay detection7- Results8- Conclusions and Future works

2

Agenda

1- Introduction2- Proposed System3- Dominant Color4- Shot-boundary detection5- Shot-type Classification6- Replay detection7- Results8- Conclusions and Future works

3

Sports Around The World

• Sports, especially soccer, attract many people.• In the past, people were watching their local

league. • Now with the evolution of communications

(Satellite, Internet … etc) they can watch more than one league at the same time.

4

Agenda

1- Introduction2- The Proposed System3- Dominant Color4- Shot-boundary detection5- Shot-type Classification6- Replay detection7- Results8- Conclusions and Future works

5

The Proposed System

6

Agenda

1- Introduction2- Proposed System3- Dominant Color4- Shot-boundary detection5- Shot-type Classification6- Replay detection7- Results8- Conclusions and Future works

7

What is meant by Dominant Color?

• The dominant color is the color that is filling most of the given area, and it is different between various fields.

• The field of any sport must contain a unique dominant color, which can be used to differentiate among various sports by detecting their field.

8

Different Dominant Color Fields

9

Agenda

1- Introduction2- Proposed System3- Dominant Color4- Shot-boundary detection5- Shot-type Classification6- Replay detection7- Results8- Conclusions and Future works

10

What is meant by Shot ?

Separated view comes from multiple cameras views that are positioned at different locations along the pitch.

11

Camera 1

Camera 1

Correlation Between Frames

• The correlation between frames in the same shot is a very important indicator to detect the similarity between them.

• When there is noticeable change we can conclude the starting of new shot.

12

Shot-boundary detection algorithm

For the proposed system, three features have been used for shot-boundary detection in sports video:

• The difference in color histogram similarity.• The motion difference.• Macro blocks change.

13

Instant Transition (cut)

14New shot

Gradual Transition

15

Comparing a range of placement values instead of one placement value in the given frames.

New shot

Agenda

1- Introduction2- Proposed System3- Dominant Color4- Shot-boundary detection5- Shot-type Classification6- Replay detection7- Results8- Conclusions and Future works

16

Shot Classification

Different shot types:

Long shot: A long shot displays the global view of the field; long shots almost always display some part of the stadium.

Medium (In-field) shot: A medium shot, where a whole human body is usually visible, is a zoomed-in view of a specific part of the field.

Close-up Shot: A close-up shot usually shows the above-waist view of a player or referee.

Audience (Out-of-field) Shot: The audience, coach, and other shots are denoted as out-of-field shots.

17

Shot Classification using Grass Ratio

18

Play / Break

19

Agenda

1- Introduction2- Proposed System3- Dominant Color4- Shot-boundary detection5- Shot-type Classification6- Replay detection7- Results8- Conclusions and Future works

20

Importance of Replay Detection

Replay Detection

Summarization Cinematic Features

21

• Replay is a video editing way, and it is often used to emphasize an important segment.

• Replay segments are good indicators for exciting events in any sport video.

Logo Based Replay

Logo frame contains a large contrast object and is usually animated within 10-20 frames with a general pattern of “smallest–biggest–smallest”. 22

Logo detection Algorithm

23

Agenda

1- Introduction2- Proposed System3- Dominant Color4- Shot-boundary detection5- Shot-type Classification6- Replay detection7- Results8- Conclusions and Future works

24

Results

25

• The proposed system was evaluated using videos for soccer matches of five international soccer championships.

• The proposed system performs very well as its analysis results achieve high accuracy.

• Experiments show that the system has attained very high precision and reasonable recall ratios.

Results

26

Results

27

Agenda

1- Introduction2- Proposed System3- Dominant Color4- Shot-boundary detection5- Shot-type Classification6- Replay detection7- Results8- Conclusions and Future works

28

Conclusions• We introduced the effectiveness and the efficiency of all applied methods

in our proposed system such as :1-Dominant Color Detection.2-Shot-boundary detection.3-Shot-type classification.4-Replay detection.

• The proposed system was evaluated on a variety of soccer matches and demonstrated that it is capable of achieving good performance characterized by high recall and precision against a manually defined ground truth.

29

Future works

• Increasing the number of soccer videos and championships being examined.

• Apply different machine learning approaches to differentiate events from one championship to another.

• Moreover, additional phases may be added in order to extend the proposed system for generating a summarized version soccer videos and highlighting the most important events during the match.

30

© 2000, Cisco Systems, Inc. www.cisco.com 8-31

Thank You !Thank You !

31