Find here all projects I did during my studies:
-
Automated Camera Stabilization and Calibration for Intelligent Transportation Systems:
These projects were developed as part of the Providentia++ research project during my Guided Research at the Chair of Robotics, Artificial Intelligence, and Real-Time Systems. Later, we documented our work in our paper An Online Self-Correcting Calibration Architecture for Multi-Camera Traffic Localization Infrastructure and successfully submitted it at IEEE Intelligent Vehicles Symposium.- Static Calibration focuses on the static calibration of sensor data, particularly to improve positioning accuracy in autonomous driving systems.
- Dynamic Stabilization deals with stabilizing sensor measurements in dynamic environments to ensure consistent and precise data acquisition.
-
OpenDRIVE:
A side project of my Guided Research, implementing a parser for the OpenDRIVE 1.6 standard.- OpenDRIVE is a widely used format for describing road networks and environmental data in autonomous driving simulations.
- The parser extracts landmarks and lanes from OpenDRIVE files for use in the calibration process.
-
Volumetric-Fusion:
In this practical project, I developed a complete 3D reconstruction pipeline using a multi-camera setup.- The pipeline includes all steps from image preprocessing and 3D registration of cameras to shape reconstruction using a voxel grid approach.
- Volumetric fusion techniques are widely used in computer vision for applications such as 3D scanning, augmented reality, and robotics.
-
GAIL 4 BARK:
During an internship at fortiss, I implemented Generative Adversarial Imitation Learning (GAIL) agents for the BARK simulator.- GAIL is an imitation learning method that allows agents to learn behavior strategies by observing expert demonstrations.
- The BARK simulator is a platform for autonomous vehicle simulations, used for research and validation of control algorithms.
-
Generalization-Ideas-in-Deep-Learning:
In this seminar, I explored different measures for the generalization ability of neural networks.- Generalization refers to how well a trained model can handle new, unseen data without overfitting.
- I analyzed various techniques to improve generalization, including regularization methods and architectural designs.
-
Capstone-Project:
My first semester project and overall my very first project: A terminal-based maze game with procedurally generated levels.- Features include enemies, traps, keys, and doors, along with a side-scrolling mechanic controlled via arrow keys.
- This project served as an introduction to game development and algorithm design, particularly in procedural generation.