Similarly to most of the existing tracking algorithms, the IPAN Tracker assumes that a point can only belong to a single trajectory. An important additional assumption is that an upper estimate for the maximum speed, , is given. Currently, the duration of an occlusion is limited to 2 frames, but this restriction can be lifted.
The algorithm is based on the idea of competing trajectories. Correspondences to the previous and the subsequent frames are established as the result of a competitive hypothesis testing process similar to the one applied in the original two-frame matching procedure [2]. The basic difference is that three consecutive frames are now used to obtain the two hypothetic displacement vectors needed to form a smoothness-based cost function.
The matching algorithm has the following three steps: the initialization, the processing of the subsequent frames, and the post-processing. The initialization procedure operates on the first three frames and induces the tracking process. The points in are linked to the corresponding points in and . Starting from , a different matching procedure is applied. When all frames have been matched, a post-processing procedure is used to reconsider the points that temporarily disappeared and later re-appeared. This procedure attempts connecting the corresponding endpoints of the broken trajectories. The three steps are described below in more detail.