Deep multi-object tracking for ground truth trajectory estimation

Important results from the project

The aim of this project was to develop algorithms that will provide high-precision estimates of the trajectories of all dynamic objects in the vicinity of the host vehicle. The purpose was to obtain an efficient technique to extract estimates that can be viewed as ground truth. From an academic perspective, the goal was to employ a PhD student who conducts research related to above goals and concludes his studies successfully. The mentioned goals were mostly fulfilled through the research carried out by a PhD student at Chalmers university. The student finished his studies successfully.

Expected long term effects

This project has had significant academic contributions. The research conducted during the course of this project has helped in establishing a theoretical framework for trajectory estimation of multiple dynamical objects. In addition, this project has generated several ideas regarding the use of deep learning methods for tracking. These ideas have been the basis of many follow up PhD projects. Within the course of this project in total, 6 master theses, one Licentiate thesis, one PhD thesis, 15 journal papers and 7 conference papers have been written.

Approach and implementation

The project was coordinated by Zenuity AB from the beginning and was taken over by Zenseact AB from July 2020. The coordinator at Zenuity and Zenseact was Dr. Daniel Svensson and later Dr. Maryam Fatemi became the coordinator for the project. The Majority of the work and research was carried out by Yuxuan Xia, a PhD student who was supervised by Prof. Lennart Svensson at Chalmers university. Dr. Karl Granström, who was with Chalmers at the beginning of this project, also contributed to it. Yuxuan Xia, finished his studies successfully and made significant contributions along his journey.