Date : 29/12/2025
Hour:16.30
Student Name & ID: Erdem Abdullah Eser & 235105108
Supervisor: Dr. Öğr. Üyesi GÖKHAN KORAY GÜLTEKİN
Topic: Registration of Event Camera Data
Link or Room: Dekanlık Toplantı Odası (Face to Face)
Abstract:
In this study, we aim to adapt the image registration techniques commonly used in conventional frame-based cameras to the domain of event-based vision sensors. Traditional registration approaches rely on detecting prominent image features in consecutive frames and aligning multiple images by matching these features.
Event cameras, however, operate fundamentally differently. Instead of producing images at fixed intervals, they output a continuous stream of events, each containing spatial coordinates (x, y), timestamp (t), and polarity. Due to the absence of intensity frames, classical registration methods cannot be applied directly and require alternative representations or new algorithmic formulations.
Our work focuses on two main approaches:
Converting event data into frame-like representations—such as event histograms, time surfaces, or simple polarity-based visualizations—and then applying conventional feature-based registration techniques on these reconstructed frames.
Performing registration directly on the event stream without generating frames. In this approach, events are aligned using their spatial, temporal, and polarity information to estimate geometric transformations purely in the event domain.
By comparing these two strategies, we investigate their strengths and limitations, and explore methods for improving registration performance in event-based perception systems. This study aims to contribute to a deeper understanding of how registration can be effectively achieved using event camera data.