Only IP97 can handle all events: occlusions, entries and extis. The other four algorithms have different sets of admissible events. The intersection of these sets is empty. (See table 1.) In our tests, each algorithm is compared, under its own conditions, to IP97.
Two essentially different data sets are used in the evaluation.
The sequences synthesized for SJ87 and RS91
contain no motion across image border, while the sequences generated
for HW89, SS90 and IP97 contain border crossings. (For HW89, entry is only allowed.)
It is important to emphasize that
the results for these two data sets should not be compared directly.
The reason is that the same number of trajectories in the two data sets
does not mean the same average point density:
when the points are allowed to exit and enter, many trajectories are short.
When no motion across border is allowed, a generated trajectory is only
accepted if it extends through the whole sequence, with a few possible occlusions.
This higher average point density leads to more frequent groupings and ambiguities.
The difference in density for the same 
grows with the mean speed.