As discussed during the user conference earlier, this PR is created to demonstrate an example implementation of JIPDA.
The implemented JIPDA is an implementation of the equations specified in Section 3 of [1]. Note that no consideration of track-birth is taken, since this is handled separately by the PHD filter described in Section 4 of the same paper.
A relation between JPDA and JIPDA is also drawn in Section 4.4 of [2], where I also investigated how we can use JPDA (as well as algorithms such as EHM) as a black box to obtain an implementation of JIPDA. This is also the approach used in the implementation shown in this PR.
Some notes:
The added components assume that Tracks have an "exist_prob" property. This is added as a dynamic property at runtime (see here).
A JIPDA data associator has been implemented. It's operation is identical to JPDA, except from the addition of lines 210-212 and 238-239.
An IPDA hypothesiser has been created that performs prediction of the existence variable:
Overall, its operation is identical to the PDA hypothesiser, except from the addition of lines 218-222 and a slight modification of the probability calculation in line 262.
The hypothesiser also pre-caclculates the value of a weight w, required to adjust the null hypothesis weight computed by JIPDA (see here). An explanation of why this is necessary can be found in Section 4.4.5 of [2].
An additional metadata property has been added to SingleHypothesis to enable us to communicate the value of w to JIPDA.
References:
[1] Horridge, Paul & Maskell, Simon & Ltd, Qinetiq. (2011). Using a Probabilistic Hypothesis Density filter to confirm tracks in a multi-target environment.
[2] Vladimirov, Lyudmil (2021) Mathematical Models and Monte-Carlo Algorithms for Improved Detection of Targets in the Commercial Maritime Domain. Doctor of Philosophy thesis, University of Liverpool.
As discussed during the user conference earlier, this PR is created to demonstrate an example implementation of JIPDA.
The implemented JIPDA is an implementation of the equations specified in Section 3 of [1]. Note that no consideration of track-birth is taken, since this is handled separately by the PHD filter described in Section 4 of the same paper.
A relation between JPDA and JIPDA is also drawn in Section 4.4 of [2], where I also investigated how we can use JPDA (as well as algorithms such as EHM) as a black box to obtain an implementation of JIPDA. This is also the approach used in the implementation shown in this PR.
Some notes:
w
, required to adjust the null hypothesis weight computed by JIPDA (see here). An explanation of why this is necessary can be found in Section 4.4.5 of [2].metadata
property has been added toSingleHypothesis
to enable us to communicate the value ofw
to JIPDA.References: [1] Horridge, Paul & Maskell, Simon & Ltd, Qinetiq. (2011). Using a Probabilistic Hypothesis Density filter to confirm tracks in a multi-target environment. [2] Vladimirov, Lyudmil (2021) Mathematical Models and Monte-Carlo Algorithms for Improved Detection of Targets in the Commercial Maritime Domain. Doctor of Philosophy thesis, University of Liverpool.