automatic detection of highlights from a cricket match

32
Outline The Story of 641.com Machine Learning Lab 1 Computer Vision Lab 2 1 Department of Computer Science Indian Institiue of Science 2 Department of Electrical Engineering Indian Institiue of Science 31-July-2008 ML lab & CV Lab The Story of 641.com

Upload: csk-varma

Post on 15-May-2015

360 views

Category:

Technology


4 download

DESCRIPTION

This is a presentation of work done by Chekuri Srikanth Varma, Sreedal Menon, Dinesh Govindarajan at Machine Learning Lab, IISc in 2008 as a sponsored project with AOL India

TRANSCRIPT

Page 1: Automatic detection of highlights from a Cricket Match

Outline

The Story of 641.com

Machine Learning Lab1 Computer Vision Lab2

1Department of Computer ScienceIndian Institiue of Science

2Department of Electrical EngineeringIndian Institiue of Science

31-July-2008

ML lab & CV Lab The Story of 641.com

Page 2: Automatic detection of highlights from a Cricket Match

Outline

Outline

1 Runs, Runs everywhere

2 Where is the problem?

3 So What?

4 How to do that?

5 Ok.. How good is it?

6 What now?

7 Lets see it !

ML lab & CV Lab The Story of 641.com

Page 3: Automatic detection of highlights from a Cricket Match

Outline

Outline

1 Runs, Runs everywhere

2 Where is the problem?

3 So What?

4 How to do that?

5 Ok.. How good is it?

6 What now?

7 Lets see it !

ML lab & CV Lab The Story of 641.com

Page 4: Automatic detection of highlights from a Cricket Match

Outline

Outline

1 Runs, Runs everywhere

2 Where is the problem?

3 So What?

4 How to do that?

5 Ok.. How good is it?

6 What now?

7 Lets see it !

ML lab & CV Lab The Story of 641.com

Page 5: Automatic detection of highlights from a Cricket Match

Outline

Outline

1 Runs, Runs everywhere

2 Where is the problem?

3 So What?

4 How to do that?

5 Ok.. How good is it?

6 What now?

7 Lets see it !

ML lab & CV Lab The Story of 641.com

Page 6: Automatic detection of highlights from a Cricket Match

Outline

Outline

1 Runs, Runs everywhere

2 Where is the problem?

3 So What?

4 How to do that?

5 Ok.. How good is it?

6 What now?

7 Lets see it !

ML lab & CV Lab The Story of 641.com

Page 7: Automatic detection of highlights from a Cricket Match

Outline

Outline

1 Runs, Runs everywhere

2 Where is the problem?

3 So What?

4 How to do that?

5 Ok.. How good is it?

6 What now?

7 Lets see it !

ML lab & CV Lab The Story of 641.com

Page 8: Automatic detection of highlights from a Cricket Match

Outline

Outline

1 Runs, Runs everywhere

2 Where is the problem?

3 So What?

4 How to do that?

5 Ok.. How good is it?

6 What now?

7 Lets see it !

ML lab & CV Lab The Story of 641.com

Page 9: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Runs, Runs everywhere

Cricket is all about ‘Runs’ & ‘Wickets’

30 Years of Runs and Wickets in archives

Atleast 50 International Matches per player per year

On the average 4-5 fours+sixes+wickets per match perplayer

12-14 players per international team

Over 10 international teams

So what

Over 10,50,000 interest points!

ML lab & CV Lab The Story of 641.com

Page 10: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Can we extract & classify 6,4 & W automatically

In Live video streams?

Can we do it in real time?

How accurately can we do it?

In Archived videos?

How good is the summary?How much ‘valid’ data is lost?How much compression is obtained?

Retrieval speed, Preprocessing etc..

ML lab & CV Lab The Story of 641.com

Page 11: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Implications

1 Automated highlights in real time.2 Live streaming of highlights to Mobiles (even in low

bandwidth environments)3 Interactive search platform for yesteryears great

performances

ML lab & CV Lab The Story of 641.com

Page 12: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Implications

1 Automated highlights in real time.2 Live streaming of highlights to Mobiles (even in low

bandwidth environments)3 Interactive search platform for yesteryears great

performances

ML lab & CV Lab The Story of 641.com

Page 13: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Implications

1 Automated highlights in real time.2 Live streaming of highlights to Mobiles (even in low

bandwidth environments)3 Interactive search platform for yesteryears great

performances

ML lab & CV Lab The Story of 641.com

Page 14: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Implications

1 Automated highlights in real time.2 Live streaming of highlights to Mobiles (even in low

bandwidth environments)3 Interactive search platform for yesteryears great

performances

ML lab & CV Lab The Story of 641.com

Page 15: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

How to do that?

ML lab & CV Lab The Story of 641.com

Page 16: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

How to do that?(contd...)

ML lab & CV Lab The Story of 641.com

Page 17: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

How to do that?(contd...)

ML lab & CV Lab The Story of 641.com

Page 18: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Ok.. How good is it?

System Specs..

AMD turion 1.7Ghz, 512 MB RAM , Matlab 7.3 (8x1.25hrvideos)

1 AccuracyAvg. Misclassifaction rate : 5% (May miss some scorechanges)

2 How long will it take !!Complete run (2 hr video ) : 25 min

Human intervention?

Completely Automated!!!

ML lab & CV Lab The Story of 641.com

Page 19: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Ok.. How good is it?

System Specs..

AMD turion 1.7Ghz, 512 MB RAM , Matlab 7.3 (8x1.25hrvideos)

1 AccuracyAvg. Misclassifaction rate : 5% (May miss some scorechanges)

2 How long will it take !!Complete run (2 hr video ) : 25 min

Human intervention?

Completely Automated!!!

ML lab & CV Lab The Story of 641.com

Page 20: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Ok.. How good is it?

System Specs..

AMD turion 1.7Ghz, 512 MB RAM , Matlab 7.3 (8x1.25hrvideos)

1 AccuracyAvg. Misclassifaction rate : 5% (May miss some scorechanges)

2 How long will it take !!Complete run (2 hr video ) : 25 min

Human intervention?

Completely Automated!!!

ML lab & CV Lab The Story of 641.com

Page 21: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Ok.. How good is it?

System Specs..

AMD turion 1.7Ghz, 512 MB RAM , Matlab 7.3 (8x1.25hrvideos)

1 AccuracyAvg. Misclassifaction rate : 5% (May miss some scorechanges)

2 How long will it take !!Complete run (2 hr video ) : 25 min

Human intervention?

Completely Automated!!!

ML lab & CV Lab The Story of 641.com

Page 22: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Ok.. How good is it?

System Specs..

AMD turion 1.7Ghz, 512 MB RAM , Matlab 7.3 (8x1.25hrvideos)

1 AccuracyAvg. Misclassifaction rate : 5% (May miss some scorechanges)

2 How long will it take !!Complete run (2 hr video ) : 25 min

Human intervention?

Completely Automated!!!

ML lab & CV Lab The Story of 641.com

Page 23: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

What now?

1 Live Mobile streaming of highlights.2 From TV broadcast to Mobile streaming : under 2 min.3 Easy to archive and retrieve old videos.4 can be used for pictorial summarization.

Lighter Note

Seems like a good startup idea!!

ML lab & CV Lab The Story of 641.com

Page 24: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Research Scope

1 Action Recognition (like hooks ,pulls etc.,).2 Player detection and tracking .

Generic

Can we make some of these techniques work for other sportslike football.

ML lab & CV Lab The Story of 641.com

Page 25: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Research Scope

1 Action Recognition (like hooks ,pulls etc.,).2 Player detection and tracking .

Generic

Can we make some of these techniques work for other sportslike football.

ML lab & CV Lab The Story of 641.com

Page 26: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Research Scope

1 Action Recognition (like hooks ,pulls etc.,).2 Player detection and tracking .

Generic

Can we make some of these techniques work for other sportslike football.

ML lab & CV Lab The Story of 641.com

Page 27: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Lets see it !

1 Streaming through web site.2 Streaming to mobile phones.3 Real time streaming of live broadcast to mobile phones.

ML lab & CV Lab The Story of 641.com

Page 28: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Lets see it !

1 Streaming through web site.2 Streaming to mobile phones.3 Real time streaming of live broadcast to mobile phones.

ML lab & CV Lab The Story of 641.com

Page 29: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Lets see it !

1 Streaming through web site.2 Streaming to mobile phones.3 Real time streaming of live broadcast to mobile phones.

ML lab & CV Lab The Story of 641.com

Page 30: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

More Statistics?

ML lab & CV Lab The Story of 641.com

Page 31: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Semi Automated Approach

ML lab & CV Lab The Story of 641.com

Page 32: Automatic detection of highlights from a Cricket Match

Runs, Runs everywhereWhere is the problem?

So What?How?

How good is it?What now?Lets see it !

Semi Automated Approach - Statistics

Even though the Completely automated algorithm has realgood results, it misses some shots.

Reason: The scorecard boundary is not perfect.

Adhoc Solution: Semi Automated approach in which thescorecard boundary is manually input by the user.

About 10% improvement over and above the preformanceof the CA algorithm is obtained.

ML lab & CV Lab The Story of 641.com