Alexander Greger och Fabian Burman, Elektroteknik

​Titel: Flexible road modelling using a multi-object approach
​Examinator: Lars Hammarstrand

Abstract:
Advanced Driver Assistance Systems help drivers by increasing traffic safety and comfort. Some of these systems, such as Adaptive Cruise Control, are dependent on an accurate representation of the road in front of the vehicle when making decisions. If the estimated geometry is inaccurate, the systems' decisions will be based on a false representation of reality, which can lead to accidents. By improving the road geometry estimation utilized by such systems, the safety of all travelers on the road can be increased.

A problem present in many modern road estimation methods is that they are inaccurate when modelling diverging or non-parallel lanes. This thesis presents a set of flexible yet robust algorithms for estimating the host lane and any diverging lanes. The problem is approached as a multi-object tracking problem, and the thesis examines two different multi-object tracking filters, the Gaussian Mixture Probability Hypothesis Density (GM-PHD) filter and the Poisson Multi Bernoulli Mixture (PMBM) filter. The PMBM is a robust tracker that performs well in complex and easy scenarios, whereas the GM-PHD filter is a simpler and more computationally efficient tracker that performs well on less complex tracking scenarios.

The thesis results show that both algorithms can model the host lane and diverging lanes on a highway. The two filters have similar performance when modelling the host lane, but the Poisson Multi Bernoulli Mixture is more proficient at detecting the diverging lanes. However, a different evaluation method is required to properly assess the geometry of the diverging lanes, since the current method only evaluates the detection frequency.

Welcome!
Alexander, Fabian and Lars
Kategori Studentarbete
Plats: EDIT-rummet, meeting room, Hörsalsvägen 11, EDIT trappa C, D och H
Tid: 2022-06-07 15:00
Sluttid: 2022-06-07 16:00

Sidansvarig Publicerad: on 25 maj 2022.