Deep multi-object tracking for self-driving vehicles

Research Project, 2021 – 2025

This project aims to investigate and develop deep learning-based multi-object tracking (MOT) algorithms for different sensors, such as radar, camera and lidar. The goal of the project is to push the current frontier for a variety of problems that are of key importance to robotics in general and autonomous vehicles in particular. The project has the following two main objectives:

-     Develop deep MOT algorithms that provide state-of-the-art performance in a model-based setting, i.e., in a setting where traditional methods are currently the default solution.

-     Identify a deep learning architecture for model-free MOT that can make optimal use of real radar, camera and lidar data.

In the first work package, we will develop deep learning models that can replace the traditional MOT algorithms in the model-based regime, which means that data can be freely generated (simulated) from our models. In the second work package, we will work with real data for self-driving vehicles.

The project is expected to deliver results and methods in object tracking and deep learning that will advance the current research front and can be integrated and used in Zenseact's software for self-driving cars.