With the arrangement you have described I would say it's very very hard to do what you want.
- You know where the robot is
- You know where the robot has to get to
- You can easily calculate the direction the robot has to go in (simple trigonometry from school)
You don't know what direction the robot is facing, so you don't know how to turn the robot to get it to drive in the direction you need.
So either you need some kind of calibration procedure whereby the computer moves the robot in different directions and watches what the IR LED does to determine what direction the robot is facing so it can move it to face the right way (tricky to do, and rather messy), or some way for the computer to visually calculate the direction of the robot.
The latter is probably the easiest. Instead of just one IR LED on the top of the robot you would need an array of IR LEDs on top of the robot forming a shape. What that shape is can be up to you, but there is one big constraint: It must have a clear direction to it (i.e., it must have no rotational symmetry. Triangles are good since they only need 3 LEDs. An equilateral triangle though wouldn't work since in three different rotations it would look the same. However, a single LED at the front of the robot, with two LEDs at the back of the robot, with the two back LEDs closer together than the distance from the front to the back, would not have any rotational symmetry.
You could then, (theoretically, assuming you have control over the camera software), get the three points from the play area and calculate the direction the robot is facing (tangent from mid-way between the two close points passing through the distant point) so you can then calculate the angle the robot would need to turn through so it is facing the direction you need it to.
Remember in math class, doing trigonometry (sin, cos, tan, pythagoras, etc) thinking "Why the hell do we need to know this stuff?!", well, now you know.