Contents
Introduction
The aim of this project is the automatic generation of Natural-Language
commentary of a robot soccer match. Overall process would involve three
parts:
-
Image processing : This would involve graphics and object recognition
of the positions of players and the ball over the period of the match.
-
Spatio-temporal reasoning (Event Recognition): This would involve
converting these position data over time to a higher level information,
here the detection of events of the soccer match.
-
Reporting : This involves converting this detected event to Natural
language, here English in such a way as to sound natural avoiding frequent
repetitions, creating different templates even for the same event so as
to make it interesting.
This project had been started by Deepayan Chakrabarti <deepay@iitk.ac.in>
and I aim to extend and improve it based on the suggestions put by Deepayan
in his report. The other details of the project can be seen from his report
web page
http://www.iitk.ac.in/mech/courses/ME768/98/
Comparison with Past Work
Deepayan has detected some basic moves which I shall call as primitives.
I would in my project include some more of these primitives and then compound
these primitives to generate higher level commentary.
Some of the events already reported by Deepayan are:
-
Tackles
-
Passes
-
Overtakes
-
Goals
-
Run along the flanks
Some primitives I plan to add are:
-
Good pass
-
Dodge or dribble
-
Strike at goal
-
Long or short pass
These primitives will be compounded to detect:
-
Good move
-
Player role : defender, striker, mid-fielder
-
Strategy (offense) based probabilistic prediction of later events
For the source code of this project CLICK HERE
This report was prepared by Samit Panjiar (96250) as a part of the
project component in the Course on Artificial Intelligence in the January-April
Semester, 2000. (Instructor : Amitabha
Mukerjee )