Gradually migrating Gaussians – A more stable product approximation CPHD filter for multisensor multitarget tracking
- Authors
- Corporate Authors
- Defence Research and Development Canada, Atlantic Research Centre, Halifax NS (CAN);McGill Univ, Montreal QUE (CAN) Dept of Electrical Engineering
- Abstract
- For the single sensor scenario, provided certain modeling assumptions are reasonable, the cardinalized probability hypothesis density (CPHD) filter is an effective and computationally tractable solution for multi-target tracking. The key assumptions are: (i) targets move and generate measurements independently; (ii) clutter is generalized Poisson and independent of the measurements; (iii) each measurement is generated by a single target or by clutter; (iv) each target generates only one measurement. The single-sensor CPHD filter has been tested extensively through numerical simulations and practical deployments. There has been much less work addressing the extension of the filter (and other filters based on random finite set theory) to the multi-sensor setting. Two major approaches have been proposed: the heuristic iterated-corrector approximation [1] and the multisensor product approximation [2]. In the iterated-corrector CPHD filter, the measurements from each sensor are processed sequentially. The output PHD obtained after processing the measurements from the first sensor is used as the input PHD for processing the measurements of the second sensor and so on. Disturbingly, the final result depends on the order in which sensor data is processed, and in some settings the choice of order can significantly change the result [3]. The product approximation CPHD filter is more principled, does not impose significantly stronger assumptions and has a similar computational overhead.
- Report Number
- DRDC-RDDC-2016-N036 — External Literature
- Date of publication
- 02 Jun 2014
- Number of Pages
- 1
- DSTKIM No
- CA043132
- CANDIS No
- 804420
- Format(s):
- Electronic Document(PDF)
Document 1 of 1
- Date modified: