Multi-target tracking involves jointly estimating the number of targets and their dynamic states from noisy sensor measurements. Notable multi-target tracking algorithms such as the multiple hypothesis filter (MHT) and joint probabilistic data association (JPDA) filter are the most popular multi-target tracking filters. However, these filters were developed as extensions to single-target tracking, and in 2017, Ronald Mahler noted that, while these filters work well, they were not developed within a rigorous mathematical framework. Mahler made a case for the development of a new multitarget tracking paradigm, the random finite set (RFS) approach, based on concepts from point process theory [13]. Since its formal inception in 2007, the RFS approach has garnered over 1000 publications from more than 20 different countries and has successfully been applied to areas such as defense, cell microscopy, robotics, and space debris tracking [11, 4]. In 2017, Vo and Vo proposed the joint delta-generalized labeled multi-Bernoulli (delta-GLMB) filter as the first multi-target tracking filter that provides an exact closed-form solution to the Bayes multi-target tracking filter (the RFS analogue to Bayes recursive filter) [24]. Over the past few years, the delta-GLMB filter [24, 22, 23] has gained much attention and is challenging the dominance of classic multi-target racking methods. The focus of this thesis is translating delta-GLMB code developed by the authors of [22] to python via an object-oriented design. A synthetic dataset was created to provide a baseline for comparison to the MHT algorithm. The MHT performance was determined using an open-source implementation on the dataset. The comparison between the MHT and delta-GLMB showed that both filters performed similarly with respect to cardinality estimation, although the delta-GLMB filter performed slightly better with respect to localization. This thesis also serves as a tutorial to introduce the reader to multi-target tracking and, specifically, to provide a practical implementation guide of the delta-GLMB filter proposed in [22]. This thesis also aims to provide minimal mathematical and theoretical details sufficient to understand RFS theory and the delta-GLMB filter
Metrics
277 File views/ downloads
178 Record Views
Details
Title
A practical implementation of the [delta]-GLMB filter
Creators
Asaf Yehuda Rothschild
Contributors
Thomas A. Chmielewski (Advisor)
Awarding Institution
Drexel University
Degree Awarded
Master of Science (M.S.)
Publisher
Drexel University; Philadelphia, Pennsylvania
Number of pages
x, 83 pages
Resource Type
Thesis
Language
English
Academic Unit
College of Engineering (1970-2026); Electrical (and Computer) Engineering [Historical]; Drexel University