A comparative analysis of object tracking methods
Kamdem Teyou, Louis Mozart (2023)
Diplomityö
2023
School of Engineering Science, Laskennallinen tekniikka
Kaikki oikeudet pidätetään.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe2023060852867
https://urn.fi/URN:NBN:fi-fe2023060852867
Tiivistelmä
Multiple object tracking consists of tracking several targets simultaneously.
This thesis focuses on demonstrating the effectiveness of random finite set-based methods in object tracking, considering both single-object and multiple-object tracking scenarios with linear and nonlinear Gaussian models. In single object tracking, we investigate various challenging scenarios such as tracking smooth targets, handling missing data, managing outliers, and tracking discontinuous targets with shaped corners. On the other hand, in multiple object tracking, we specifically address targets that appear and disappear, as well as the management of clutter within the tracking area. To evaluate the random finite set-based methods, we conduct a comprehensive step-by-step comparison against traditional approaches including the Kalman filter, extended Kalman filter, and the Rao-Blackwellized Monte Carlo data association filter. Additionally, we explore the Gaussian mixture probability hypothesis density filter and the Poisson multi-Bernoulli mixture filter as prominent random finite set methods. The evaluation results showcase the advantages of random finite set-based methods in handling diverse tracking scenarios with reduced computational time and low error rates. Challenges are encountered by the Rao-Blackwellized Monte Carlo data association and Kalman filters in tracking discontinuous targets with shaped corners. Also, the Kalman and extended Kalman filters face difficulties in the presence of outliers and missing data in the measurements.
This thesis focuses on demonstrating the effectiveness of random finite set-based methods in object tracking, considering both single-object and multiple-object tracking scenarios with linear and nonlinear Gaussian models. In single object tracking, we investigate various challenging scenarios such as tracking smooth targets, handling missing data, managing outliers, and tracking discontinuous targets with shaped corners. On the other hand, in multiple object tracking, we specifically address targets that appear and disappear, as well as the management of clutter within the tracking area. To evaluate the random finite set-based methods, we conduct a comprehensive step-by-step comparison against traditional approaches including the Kalman filter, extended Kalman filter, and the Rao-Blackwellized Monte Carlo data association filter. Additionally, we explore the Gaussian mixture probability hypothesis density filter and the Poisson multi-Bernoulli mixture filter as prominent random finite set methods. The evaluation results showcase the advantages of random finite set-based methods in handling diverse tracking scenarios with reduced computational time and low error rates. Challenges are encountered by the Rao-Blackwellized Monte Carlo data association and Kalman filters in tracking discontinuous targets with shaped corners. Also, the Kalman and extended Kalman filters face difficulties in the presence of outliers and missing data in the measurements.