Super-resolution method for reconstructing potential image based on dynamic vision sensor

Abstract

The invention discloses a super-resolution method for reconstructing a potential image based on a dynamic visual sensor, relates to the technical field of event camera application, and solves the problems that edge details of a grayscale image are not sharp enough, the grayscale of the same pixel value is not stable enough, the dynamic range is liable to lose and the like in a reconstruction junction in the prior art. The method is realized through a neural network structure, and comprises two steps of potential grey-scale map reconstruction and multi-image fusion; a high-quality high-resolution grey-scale map is reconstructed through a neural network architecture, and event signals and grey-scale map signals are considered at the same time, a series of potential grey-scale maps are reconstructed on the basis of the grey-scale map, and the pixel value of the fused high-resolution grey-scale map is stable and continuous. A plurality of potential grayscale frames are reconstructed, and the super-resolution of the APS grayscale image is realized by adopting a multi-image super-resolution method, so that the quality of super-resolution reconstruction is greatly superior to the reconstruction effect of the previous related methods. According to the method, the quality of the image super-resolution is improved by using a deep learning method.

Description

technical field [0001] The invention relates to the technical field of event camera applications, in particular to a super-resolution method for reconstructing latent images based on dynamic visual sensors. Background technique [0002] The image super-resolution technology (Super-resolution, SR) based on the event camera (Event Camera) can improve the resolution of the grayscale image captured by the event camera, so as to obtain a high-resolution grayscale image with richer details. High-resolution grayscale images can not only get better visualization effects, but also build a bridge between event cameras and high-level visual tasks, helping to improve the accuracy of high-level visual tasks. The event signal-guided grayscale high-resolution technology proposed in this patent first reconstructs multiple potential grayscale images from the input event signal, converts event information into grayscale information, and then fuses multiple grayscale images to achieve super-re...

AI Technical Summary

Problems solved by technology

[0033] The present invention provides a super-resolution method for reconstructing latent images based on a dynamic visual sensor to solve the problems of insufficient sharpness of the edge details of the grayscale image, insufficient stability of the grayscale of the same pixel value, and easy loss of dynamic range in the reconstruction process of the prior art. rate method

Images