Multiple Target Tracking

Definition

The problem of associating a set of anonymous position observations with sets of prior observations to construct the traces of several moving objects or targets. More precisely, given n _t position observations at time t and previous observations n _t−1 , n _t−2 , …, n _t−m , construct a set of traces, where each trace only contains the position observations from a single moving object. In a variation of the problem the number of actual moving objects may be greater than each n since objects may disappear and emerge during the observation interval.

Main Text

Multiple hypothesis tracking is one representative algorithm proposed by Reid [2] in 1979 to solve this problem. This algorithm uses a linear Kalman model to represent the movement behavior of each object and to filter observation noise. The algorithm operates in three steps. First it predicts a new system state (which includes predicted positions of each object based on the prior trajectory). Then it generates hypotheses...