A Systematic Parametric Campaign to Benchmark Event Cameras in Computer Vision Tasks

The dynamic vision sensor (DVS), or event camera, is emerging as a successful sensing solution for many application fields. While state-of-the-art datasets for event-based vision are well-structured and suitable for the designed goals, they often rely on simulated data or are recorded in loosely controlled conditions, thereby making it challenging to understand the sensor response to varying camera parameters and illumination conditions. To address this knowledge gap, this work introduces the JRC INVISIONS Neuromorphic Sensors Parametric Tests dataset, an extensive collection of event-based data specifically acquired in controlled scenarios that systematically vary bias settings and environmental factors, enabling rigorous evaluation of sensor performance, robustness, and artifacts under realistic conditions that existing datasets lack. The dataset is composed of 2156 scenes recorded with two different off-the-shelf event cameras, eventually paired with a frame camera across three different controlled scenarios: moving targets, mechanical vibrations, and rotation speed estimation; the inclusion of ground truth enables the evaluation of standard computer vision tasks. The proposed manuscript is complemented by an experimental analysis of sensor performance under varying speeds and illumination, event statistics, and acquisition artifacts such as event loss and motion-induced distortions due to line-based readout. The dataset is publicly available and, to the best of our knowledge, represents the first dataset of its kind in the literature, providing a valuable resource for the research community to advance the development of event-based vision systems and applications.