Super-resolution of spatiotemporal event-based image

Super-resolution (SR) is a useful technology to generate a high-resolution (HR) visual output from the low-resolution (LR) visual inputs overcoming the physical limitations of the sensors. However, SR has not been applied to enhance the resolution of spatiotemporal event streams recorded by dynamic vision sensors (DVSs). SR of DVS recording is fundamentally different from existing frame-based schemes since basically each pixel value of DVS recordings is an event sequence. Two problems have to be addressed in the SR of spatiotemporal event streams. The first one is how many events should be generated in each new pixel. The other one is how to generate the event sequence of each new pixel. In this work, we propose a two-stage spatiotemporal event stream SR scheme to solve the two problems. We use a sparse signal representation based resolution enhancement method and the theorem of simulating a nonhomogeneous Poisson process according to a specified rate function for the first and second problem, respectively. The proposed method is demonstrated through obtaining HR spatiotemporal event streams from various DVS recordings of both simple and complex scenes. The results show that not only the reconstructed frame of upscaled event streams is consistent with that of the LR DVS recording visually, but also the temporal properties of the HR event streams match that of the original input very well. This work enables many potential applications of event-based vision.