On using nearest neighbours with the probabilistic data association filter

The paper gives the state estimation equations for a probabilistic data association filter (PDAF) which is updated by a fixed number of nearest neighbours. These equations are based on the approach which includes track initiation and a nonuniform clutter model. Track initiation is accounted for by the event that a target can transition between being visible and invisible to the sensor. A visible target gives sensor returns that match the tracking filter sensor and target model while an invisible target gives no returns. This filter is referred to as the N³VPDAF, i.e., nearest neighbours with nonuniform clutter and visibility PDAF. Insight into the behaviour of the N³VPDAF is then found from the histogram of the measured volume of the region that contains the I nearest measurements. This is then followed by tests to measure the change in performance with the value of I over the range from 1 to 4. Performance is evaluated with a tracker assessment tool (TAT) that gives an overall measure of performance for 17 metrics. The metrics cover the categories of track establishment, track maintenance, track error and false tracks. From the filter formulae and these results, the advantages and disadvantages in selecting measurements based on nearest neighbours are given