Deep multi-object tracking for ground truth trajectory estimation
The development of automated vehicles is of great importance for the Swedish vehicle industry, and may lead to substantial gains for society. Accurate environment perception is essential to automated vehicles, since it enables vehicles to sense nearby objects, and estimate their positions as well as other relevant properties. The perception systems in modern vehicles make use of data from cameras, LIDAR sensors, etc., in order to obtain a detailed understanding of the current situation. However, more development is needed before these systems can robustly provide the accuracy required for a vehicle to drive autonomously in all situations. We will address certain aspects of environment perception, pertaining to tracking of multiple dynamic objects. Specifically, we aim to develop algorithms that provide high-precision estimates of the trajectories of all dynamic objects located around the host vehicle. The purpose is to obtain an efficient technique to extract estimates that can be viewed as ground truth, which is of utmost importance to have for the development and verification of both perception and control modules. We will investigate off-line techniques, and combining deep learning with sensor fusion, to enable extraction of as much information as possible from the data.
- Zenuity AB (Private, Sweden)
This page is only available in english
Full Professor, Deputy Head of Department, Electrical engineering
Tomas McKelvey is Deputy Head of the Department of Electrical Engineering, and Full Professor in the Signal processing research group.
Karl Granström, Postdoc, Electrical engineering
Karl Granström is a Postdoc in the Signal Processing group. Before staring at Chalmers he was a Postdoc at the University of Connecticut, where he worked together with Peter Willett and Yaakov...
Professor, Electrical engineering
Lennart Svensson is Professor in Signal Processing. His research interests include nonlinear filtering, multi-target tracking, deep learning, Bayesian inference and reinforcement learning.