|
Implementation and Evaluation of an Online, AI-based Pedestrian Detector for LiDAR Point Clouds
|
|
- Frameworks such as OpenPCDet enable training and evaluation of neural networks for object detection in point clouds, using public data sets. Within this Master thesis, the aforementioned framework will be adapted and re-structured to process a live recording of a running LiDAR sensor to detect pedestrians online. The developed system should be tested in a variety of scenarios to systematically evaluate the strengths and weaknesses of the detector.
- Requires knowledge of C++ and Python and experience with Linux operating systems.
|
|
|
*Algorithmic Detection of Railway Track Mileage in Video Streams from a Train's Perspective
|
|
- Description: In German railway infrastructure, all sections of track are measured and marked with hectometer signs. These also play an important role in digital railway operations, as they remain useful to communicate the position of the locomotive to maintenance and operating personnel as well as rescue services. For this purpose, a system is to be designed that recognizes these signs in video recordings, automatically recording the displayed mileage information. The system must meet various complex requirements: Its functionality must be verifiable, and algorithms ought to be resource-efficient and effective despite many challenging conditions (uneven lighting, position, and appearance). Such a framework rules out the use of machine learning solutions. In-depth understanding of computer vision algorithms and high-performant programming (with optional real-time aspects) will form the basis of this work.
- Requires knowledge of C++ or Python.
|
|
| Improving and Adapting Resource-constrained Cybersecurity Algorithms on Wireless Sensor Networks |
- The algorithms we designed for securing Controller Area Network (CAN) Buses against cyber attacks will be transferred and refined for use in Wireless Sensor Networks (WSNs). This adaptation is motivated by the critical observation that both CAN and WSN platforms are fundamentally characterized by severe resource limitations, making the application of conventional cybersecurity protocols impractical and inefficient.
- Requires knowledge of wireless communication, cybersecurity (authentication, encryption, etc.), MATLAB/Simulink and C or Python.
|
| Wi-Fi-Based Identity Recognition Using CSI-Linux Firmware |
- The goal of this thesis is to explore the use of the Linux 802.11n CSI Tool for collecting Channel State Information (CSI) and using it to recognize individuals based on their Wi-Fi signals. The student will set up the firmware on compatible Intel 5300 Wi-Fi chip, record data from different people, and train a deep learning model for identity classification. The main idea is to see how well this setup performs and whether it can provide cleaner or more stable CSI compared to what we usually get from other setups.
- Requires knowledge of python, wireless communication or signal processing, deep learning frameworks (PyTorch or TensorFlow), Linux environments and hardware setup.
|
| Impact of Transmitter–Receiver Distance on Wi-Fi CSI-Based Sensing Performance |
- This thesis investigates how the distance between Wi-Fi transmitter and receiver affects the quality and stability of Channel State Information (CSI) for classifying an individual’s identity. The student will record CSI data at different distances under the same experimental conditions and analyze how signal strength, noise, and classification accuracy change. After identifying an optimal distance range, additional experiments will be performed on different days to check the consistency of the results. The goal is to better understand how device placement influences CSI-based sensing performance and reliability.
- Requires knowledge of Python, wireless communication or signal processing and machine learning, interest in experimental setup and data-driven analysis.
|
| Wi-Fi CSI for Room Occupancy and Crowd Size Estimation |
- This thesis focuses on using Wi-Fi Channel State Information (CSI) to estimate how many people are present in a room. The idea is to analyze the changes in wireless signals caused by human presence and movement, and to see if these patterns can be used to detect and classify occupancy levels. The student will collect CSI data in different scenarios, process the signals, and experiment with models that can distinguish between empty and occupied environments or estimate small group sizes. The goal is to understand how reliable CSI-based sensing can be for simple, privacy-friendly occupancy detection.
- Requires knowledge of Python, wireless communication or signal processing, data analysis or machine learning, interest in experimental setup and data-driven analysis.
|
