Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Efficient approximations of the multi-sensor labelled multi-Bernoulli filter
Thesis (PhD)--Stellenbosch University, 2023.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | en_ZA en_ZA |
| Published: |
Stellenbosch : Stellenbosch University
2023
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867614109477896192 |
|---|---|
| access_status_str | Open Access |
| author | Robertson, Stuart |
| author2 | Du Preez, Johan |
| author_browse | Du Preez, Johan Robertson, Stuart |
| author_facet | Du Preez, Johan Robertson, Stuart |
| author_sort | Robertson, Stuart |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Thesis (PhD)--Stellenbosch University, 2023. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/127206 |
| institution | Stellenbosch University (South Africa) |
| language | en_ZA en_ZA |
| last_indexed | 2026-06-10T12:46:49.061Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2023 |
| publishDateRange | 2023 |
| publishDateSort | 2023 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/127206 Efficient approximations of the multi-sensor labelled multi-Bernoulli filter Robertson, Stuart Du Preez, Johan Van Daalen, Corne E. Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering. Tracking (Engineering) Binomial distribution Multisensor data fusion Thesis (PhD)--Stellenbosch University, 2023. ENGLISH ABSTRACT: Multi-object tracking is the process of jointly estimating an unknown, time-varying number of objects and their kinematic states using noisy sensor data. Multi-object tracking is typically viewed from a probabilistic perspective, and the number of objects, their kinematic state and identity are inferred from the data. The use of multiple sensors is common, as more data greatly reduces the uncertainty about both an object’s existence and its state. Multi-sensor multi-object tracking is inherently challenging, it has numerous practical applications, and it is an established field of study. In this dissertation, we propose three efficient approximate formulations of the multi-sensor labelled multi-Bernoulli (LMB) filter, which all allow the sensors’ measurement updates to be computed in parallel. Our first filter is based on the direct mathematical manipulation of the multi-sensor multi-object Bayes filter’s posterior distribution. Unfortunately, it requires the division of probability distributions and its extension beyond linear-Gaussian applications is not obvious. Our second filter approximates the multi-sensor multi-object Bayes filter’s posterior distribution using the geometric average of each sensor’s measurement-updated distribution. This filter can be used under non-linear conditions; however, it is not as accurate as our first filter. Our third filter is directly related to our second, and it can also be used under non-linear conditions. It approximates the posterior distribution using the arithmetic average of each sensor’s measurement-updated distribution, with a further loss in accuracy. In all cases, we approximate the LMB filter’s measurement update using loopy belief propagation (LBP). Our LBP algorithm is a reformulation of an existing model, and it is guaranteed to converge. Through experimentation, we show that this LBP algorithm is remarkably accurate and it has a low computational cost. Using this LBP algorithm, all of our multi-sensor LMB filters’ measurement updates have linear computational complexities in both the number of measurements and objects. If our filters’ measurement updates are computed in parallel, then this leads to significant reduction in their computational expense. Our proposed filters are the first parallelisable formulations of the multi-sensor LMB filter. They are of interest when tracking many objects using several sensors, where filter runtime is more important than filter accuracy. We compare our three approximate multi-sensor LMB filters to both an iterated-corrector LMB (IC-LMB) filter and a novel implementation of a multi-sensor LMB mixture (MS-LMBM) filter. Our simulations indicate that our proposed filters’ loss of accuracy compared to the IC-LMB and MS-LMBM filters is not significant. AFRIKAANS OPSOMMING: Multi-voorwerp volging is die proses om gelyktydig ’n onbekende, tydsveranderende aantal voorwerpe en hul kinematiese toestande af te skat deur gebruik te maak van ruiserige sensordata. Multi-voorwerp volging word tipies vanuit ’n probabilistiese perspektief beskou, en die aantal voorwerpe, hul kinematiese toestand en hul identiteit word uit die data afgelei. Die gebruik van veelvuldige sensors is algemeen, aangesien meer data die onsekerheid oor beide ’n voorwerp se bestaan en sy toestand aansienlik verminder. Multi-sensor multi-voorwerp volging is inherent uitdagend, dit het verskeie praktiese toepassings, en is ’n gevestigde studieveld. In hierdie proefskrif stel ons drie doeltreffende benaderde formulerings van die multi-sensor benoemde multi-Bernoulli (LMB) filter voor, wat almal toelaat dat die sensors se meetopdaterings in parallel bereken kan word. Die eerste filter is gebaseer op die direkte wiskundige manipulering van die multi-sensor multi-voorwerp Bayes filter se posterior verdeling. Ongelukkig vereis dit die deling van verdelings, en die uitbreiding daarvan verder as lineˆere-Gaussiese toepassings is nie voor die hand liggend nie. Die tweede filter benader die multi-sensor multi-voorwerp Bayes filter se posterior verdeling deur gebruik te maak van die geometriese gemiddelde van elke sensor se verdeling na die meetopdatering. Hierdie filter kan onder nie-lineˆere toestande gebruik word; dit is egter nie so akkuraat soos die eerste filter nie. Die derde filter is direk verwant aan die tweede filter, en dit kan ook onder nie-lineˆere toestande gebruik word. Dit benader die posterior verdeling deur gebruik te maak van die rekenkundige gemiddelde van elke sensor se verdeling na die meetopdatering, met ’n verdere verlies in akkuraatheid. In al drie gevalle benader ons die LMB-filter se meetopdatering deur gebruik te maak van lusagtige kennisvoortplanting (LBP). Ons LBP-algoritme is ’n herformulering van ’n bestaande model, en dit is gewaarborg om te konvergeer. Deur eksperimentering wys ons dat hierdie LBP-algoritme verbasend akkuraat is en dat dit ’n lae berekeningskoste het. Deur hierdie LBP-algoritme te gebruik, het al ons multi-sensor LMB-filters se meetopdaterings ’n lineˆere berekeningskompleksiteit in terme van beide die aantal metings asook die voorwerpe. Indien ons filters se meetopdaterings in parallel bereken word, lei dit tot aansienlike vermindering in hul berekeningskoste. Ons voorgestelde filters is die eerste paralleliseerbare formulerings van die multi-sensor LMB-filter. Dit is van belang wanneer baie voorwerpe gevolg word deur verskeie sensors te gebruik, waar die berekeningskoste van die filter belangriker is as die akkuraatheid van die filter. Ons vergelyk ons drie benaderde multi-sensor LMB-filters met beide ’n herhalende-opdatering LMB (IC-LMB) filter asook ’n nuwe implementering van ’n multi-sensor LMB-mengsel (MSLMBM) filter. Ons simulasies dui aan dat ons voorgestelde filters se verlies aan akkuraatheid in vergelyking met die IC-LMB en MS-LMBM filters nie beduidend is nie. Doctoral 2023-03-03T11:57:23Z 2023-05-18T07:09:46Z 2023-03-03T11:57:23Z 2023-05-18T07:09:46Z 2023-03 Thesis http://hdl.handle.net/10019.1/127206 en_ZA en_ZA Stellenbosch University application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Tracking (Engineering) Binomial distribution Multisensor data fusion Robertson, Stuart Efficient approximations of the multi-sensor labelled multi-Bernoulli filter |
| title | Efficient approximations of the multi-sensor labelled multi-Bernoulli filter |
| title_full | Efficient approximations of the multi-sensor labelled multi-Bernoulli filter |
| title_fullStr | Efficient approximations of the multi-sensor labelled multi-Bernoulli filter |
| title_full_unstemmed | Efficient approximations of the multi-sensor labelled multi-Bernoulli filter |
| title_short | Efficient approximations of the multi-sensor labelled multi-Bernoulli filter |
| title_sort | efficient approximations of the multi sensor labelled multi bernoulli filter |
| topic | Tracking (Engineering) Binomial distribution Multisensor data fusion |
| url | http://hdl.handle.net/10019.1/127206 |
| work_keys_str_mv | AT robertsonstuart efficientapproximationsofthemultisensorlabelledmultibernoullifilter |