Investigation of semantic LiDAR point cloud segmentation in an urban context.

Context

Platooning describes the automated driving of vehicles in a convoy, whereby the vehicles are networked with each other and observe the environment with the help of sensors. An important aspect in the process chain is the acquisition and processing of environmental data. LiDAR sensors generate three-dimensional point clouds of objects in the vehicle's environment. This data can be used, for example, to determine position or detect obstacles. An important step in the processing of LiDAR point cloud data is segmentation. This involves dividing the points in the point cloud into objects (for example, roads, buildings, vehicles) to enable better interpretation of the data. More information about the Tempus project can be found at https://tempus-muenchen.de.

Goals

• Overview of the state of the art of semantic LiDAR point cloud segmentation.
• Definition of appropriate evaluation criteria of segmentation methods
• Investigation of suitable processing pipelines for the segmentation of LiDAR point clouds
• Comparison of the researched conventional and machine learning based segmentation methods
• Evaluation and benchmarking of the different segmentation methods using simulation/real data.

Requirements

• Interest in the development of ML supported systems
• Programming skills (Python/C++)
• Reliable and independent way of working