Laufende Studien- und Abschlussarbeiten

Motivation

The cities of the future are expected to provide features that make the interaction with the urban environment richer, more efficient, and much more pleasant. Tasks such as navigation, planning, and mobility are expected to experience a technological revolution through, for instance, autonomous vehicles and interactive digital twins of different aspects of the urban environment. In order to achieve this vision, it is necessary to understand the environment, both in their static and dynamic components. One essential task towards this end is to be able to automatically recognize the objects in the environment, which in the ideal case should be made in a privacy-protective manner. One technology that allows to capture the environment in 3D while protecting privacy is LiDAR (light detection and ranging). In spite of the important progress that methods to recognize objects in LiDAR data have seen in recent years, the overall performance of such methods remains largely constrained, both in terms of accuracy and generalizability.

Goal

Devise, implement, and evaluate an approach that is able to recognize objects in LiDAR data, using spatio-temporal cues. The approach is expected to be effective and generalizable.

Requirements

• Programming experience required (Python and C++)
• Some hands-on experience in machine learning
• Eagerness to learn and work hard