Diplomarbeit/Master thesis
Supervisor: Martin Kampel
Status: open
Motivation
As part of the digitalisation and automation of public transport systems, an innovative platform is being developed to meet the specific operational needs of railway environments. The goal is to test intelligent systems under real-world operating conditions in order to enhance efficiency, safety, and service quality in train operations.
A key application scenario involves video- and sensor-based monitoring of railway passenger cabins to autonomously detect seat occupancy, analyse passenger flow, and identify critical health events. Modern AI methods are used to provide real-time insights to onboard systems, safety personnel, or logistical services.
Objective
The objective of this thesis/project is the design, development, and evaluation of an AI-powered visual recognition system for railway passenger environments that performs the following tasks:
- Autonomous detection of seat occupancy and analysis of passenger flow over time (e.g., boarding, alighting, movement patterns in aisles)
- Detection of potential medical emergencies, such as falls, collapses, or unusual inactivity that may indicate unconsciousness
The technical focus is on combining object detection models (e.g., YOLO, Mask R-CNN) with temporal behaviour analysis techniques (e.g., LSTM, 3D-CNNs, Transformers).
Tasks
- Literature review on seat occupancy detection, human activity recognition, and anomaly detection in video streams
- Selection or simulation of suitable datasets (synthetic or real-world passenger cabin footage)
- Development of models to detect persons, seated positions, and movement behaviours
- Implementation of a detection pipeline for critical behaviours (e.g., falls, fainting, prolonged inactivity)
- Evaluation of the system in terms of accuracy, response time, and robustness under varying lighting and camera conditions
- (Optional) Demonstration on embedded systems (e.g., NVIDIA Jetson, Raspberry Pi)
Required Skills
- Solid programming skills in Python
- Experience with computer vision frameworks (e.g., OpenCV, PyTorch, TensorFlow)
- Basic understanding of deep learning and human behaviour modelling
- Interest in safety-critical applications in public spaces
Contact: martin.kampel@tuwien.ac.at