Time-Interval-Guided Event Representation for Scene Understanding.

The recovery of scenes under extreme lighting conditions is pivotal for effective image analysis and feature detection. Traditional cameras face challenges with low dynamic range and limited spectral response in such scenarios. In this paper, we advocate for the adoption of event cameras to reconstruct static scenes, particularly those in low illumination. We introduce a new method to elucidate the phenomenon where event cameras continue to generate events even in the absence of brightness changes, highlighting the crucial role played by noise in this process. Furthermore, we substantiate that events predominantly occur in pairs and establish a correlation between the time interval of event pairs and the relative light intensity of the scene. A key contribution of our work is the proposal of an innovative method to convert sparse event streams into dense intensity frames without dependence on any active light source or motion, achieving the static imaging of event cameras. This method expands the application of event cameras in static vision fields such as HDR imaging and leads to a practical application. The feasibility of our method was demonstrated through multiple experiments.