Some subjects to cover for a Smartphone/video velocity measuring capability-
1) Jello Effect Distortions
My understanding subject to correction - CMOS imagers read out lines sequentially, for example, first at the top to last at the bottom of the frame. A single video line is read out simultaneously but there is a very slight readout delay until the next line is readout. This delay can add up to cause errors when the ball does not travel parallel to a line of video. Balls traveling parallel to lines have minimum time error and those traveling across video lines have the greatest error.
The HTC One X
Smartphone does high speed video at 120 fps. Here is an example of Jello effect distortion-
2) Alignment Error.
Objects or balls that travel away or toward the camera (instead of just across the frame, etc.) will introduce errors. An error analysis should show how sensitive the velocity measurement would be to ball travel direction alignment error. The analysis might indicate that alignment error would still yields a small velocity error, 5-10%?, for reasonable 'eyeball" alignments. Also, I believe the error of poor alignment might always yield a slower velocity that the real serve velocity(a good thing for poor alignment & bragging rights).
3) For the tennis serve placing the camera on the singles or doubles sideline and into the court, say, 8-10 feet might give a pretty good alignment viewing perpendicular to most serve trajectories.