Benjamin said:
finally, the system we built assumes the ball is being kicked from the official penalty kick point (11 meters away). the only reason we did this was because it simplifies calculating the balls trajectory (it was only a 6 month project). it would not be very difficult to modify to work with all balls kicked from surrounding locations.

I guess that will be your next challenge then!

1. http://www.youtube.com/watch?v=fLpqxJOl1K8
2. http://www.youtube.com/watch?v=Xez8y_wvi1M
3. http://www.youtube.com/watch?v=N1cvdfAJEEM
4. http://www.youtube.com/watch?v=qkR8KYzz42M

The cameras are indeed fixed around the field but this does not limit the system in any way. It is true that the system we built will not work if there are players in the cameras’ field of view but this is simply because the cameras are only black and white cameras. by using color cameras one could work around this by using, for example, an orange ball.

with regards to curved balls…the system does account for these shots as well and can easily save curled shots as well. balls with a high degree of rotation and kicked at speeds > 100km/hr into a corner have a high probability of scoring though. this is mainly because we only used 3 cameras and the visibility in the corners is somewhat limited.

finally, the system we built assumes the ball is being kicked from the official penalty kick point (11 meters away). the only reason we did this was because it simplifies calculating the balls trajectory (it was only a 6 month project). it would not be very difficult to modify to work with all balls kicked from surrounding locations.

SpiralArchitect: Cheap shot…

Ahmed I’m afraid not, at least not yet. From what I gathered on the official website and the distance the shots were taken, several conditions need to be satisfied for GoalIAS to work.

First, there absolutely can’t be any player interfering with the viewing path of the videocameras. That is, except for the ball and the shooter no one can be in close proximity to the goal or anywhere between the shooter and the “keeper”, something which would be impossible to achieve in a real-time match.

Secondly, the cameras are “fixed”. That is, they are not mobile, and they always point in the same direction. As such, the shots probably have to be taken from the same distance every time. Shoot from a little bit farther away, and the processor probably will no longer be able to compute the ball trajectory. Not to mention that any “curling” shots would be impossible to trace…

The May 22 test vs. the German Women’s National team is probably the same that Mario Gómez did: take a series of shots from a fixed position.

As for using this as a training tool for young kids… well, aside from the practical applications and everything (cost, maintenance, etc.) what would be the interest of facing an almost infallible goalkeeping machine (other than mere curiosity)? Teach the kids how to curl the ball in order to “outsmart” the robot? I fail to see the usefulness. Football is a game played against human opponents, with all the qualities and “defects” that humans bring. Learning to play the game also involves trying to exploit those defects, like fakes and dummies to send the keeper the wrong way on one-on-ones for example. You can’t really learn that from a machine…

Because if this works out, and if the robo-keeper can ‘learn’ from watching players and from previous games, then this could be a great training tool for young kids.