As target tracking is arousing more and more interest, the necessity to reliably assess tracking algorithms
in any conditions is becoming essential. The evaluation of such algorithms requires a database of sequences
representative of the whole range of conditions in which the tracking system is likely to operate, together
with its associated ground truth. However, building such a database with real sequences, and collecting the
associated ground truth appears to be hardly possible and very time-consuming.
Therefore, more and more often, synthetic sequences are generated by complex and heavy simulation
platforms to evaluate the performance of tracking algorithms. Some methods have also been proposed using
simple synthetic sequences generated without such complex simulation platforms. These sequences are
generated from a finite number of discriminating parameters, and are statistically representative, as regards
these parameters, of real sequences. They are very simple and not photorealistic, but can be reliably used
for low-level tracking algorithms evaluation in any operating conditions.
The aim of this paper is to assess the reliability of these non-photorealistic synthetic sequences for evaluation
of tracking systems on complex-textured objects, and to show how the number of parameters can be
increased to synthesize more elaborated scenes and deal with more complex dynamics, including occlusions
and three-dimensional deformations.