Development of a Deep Learning Object Detection Algorithm for Autonomous Vehicles

Lehrstuhl für Fahrzeugtechnik
Masterarbeit /
experimentell / theoretisch /  

Perception is a crucial task for autonomous vehicles. For the detection of objects in the proximity of autonomous vehicles, nowadays neural networks play an important role. For supervised training, these require large labelled datasets, which are costly to obtain. Data generated in simulation can be a viable alternative, but neural networks trained with syntethic data won't achieve similar performance as networks trained with real-world data. The scope of this thesis is the development of a neural network capable of transferring syntethic data generated in simulations to real-world data. Networks transferring data from one domain (simulation) to another domain (real-world) already exist for 2D images, e.g. CycleGAN. Instead of 2D images, 3D point clouds from LiDAR sensors are the focus of this thesis. The main challenge is the encoding, adaption and decoding of the unordered, unstructed point clouds.

  • Programming experience (Python)
  • Ideally experience in working with neural networks and deep learning frameworks (pytorch/tensorflow)
  • Involved working attitude
Verwendete Technologien
Python, Neural Networks, Deep Learning, Object Detection, Git
FTM Studienarbeit, FTM IVS, FTM Huch, FTM Informatik
Möglicher Beginn
M. Sc. Sebastian Huch