Date : 14/01/2026
Hour: 15:30
Student Name & ID: Beyza VARDAR & 245105121
Supervisor: Dr. Öğr. Üyesi GÖKHAN KORAY GÜLTEKİN
Topic: Lens Distortion Correction for Event Cameras Using Linear Scene Structures
Link or Room: Dekanlık Toplantı Odası (Face to Face)
Abstract:
Event cameras have gained significant traction in robotics and computer vision due to their high dynamic range and microsecond-level temporal resolution. However, radial distortion introduced by optical lenses remains a critical challenge, compromising geometric accuracy in applications such as SLAM and visual odometry. Addressing the limitations of standard calibration targets, which require specific motion or blinking patterns to stimulate event sensors, this study proposes a fully automated targetless calibration framework capable of handling varying fields of view. We utilize natural linear structures found in man-made environments as geometric references. The raw asynchronous event stream is transformed into structurally preserved edge maps using a temporal accumulation approach. Subsequently, an iterative energy minimization algorithm, initialized via an extended Hough transform, optimizes the distortion parameters. We evaluate both Polynomial and Division models on real-world data recorded with a DVXplorer camera using both narrow and wide-angle lenses. Experimental results demonstrate that the proposed method effectively rectifies radial distortion across different optical configurations, achieving sub-pixel Average Squared Error (ASE) and proving its robustness without the need for dedicated calibration objects.
Index Terms: Event-based vision, event camera calibration, radial distortion correction, targetless calibration, line-based optimization, wide-angle lenses