Generating Moving Point Sets

A motion model called Point Set Motion Generator (PSMG) is used to generate synthetic motion sequences for testing the tracking algorithms. In this section, we overview and justify the PSMG.

In this motion model, randomly generated points move independently. The advantage of this solution is that different types of events and trajectory ambiguities appear with statistically controllable frequencies. In fact, the PSMG is a generator of these events and ambiguities as disturbances which efficiently `test' the algorithms.

Later, simulations of assorted coherent motions will be added to test the capability to cope with correlated trajectories. Real motion sequences and detected feature points will also be considered. We believe that the disturbances that make the algorithms err will essentially remain the same. However, we understand that the occurrence probabilities of various disturbances depend on the test data. Since the algorithms may be sensitive to different disturbances in different ways, one should be cautious when interpreting the error rates presented in alternative experimental studies, including ours. Keeping this in mind, we still believe it is possible to assess the performances of the tracking techniques using the PSMG.

Each test sequence has $M= 20$ frames. The size of a frame is $200 \times 200$. In $F_{1}$, initial points are located randomly; random initial motion directions and Gaussian-distributed speeds with the mean $v$ are assigned to them. In the subsequent frames, a limited Gaussian perturbation of the velocity vectors is introduced. The speeds are limited by $v_{max}= 2v$. The probability and the maximum duration of occlusions are specified. (The results shown in section 5.3 were obtained for single-frame occlusions.) Motion across the image borders is generated by applying the PSMG to a large area enclosing the view field.

Synthetic motion created by the PSMG is exemplified below.

Click here to test the PSMG and the tracking algorithms.