Video stabilization method based on iterative strategy of recurrent neural network

Abstract

The invention discloses a video stabilization method based on a iterative strategy of a recurrent neural network. The method comprises the following steps: capturing paired video data by using a jitter video acquisition and stabilization processing hardware device; preprocessing an acquired video stream sample; designing and constructing an end-to-end full convolutional deep neural network based on an intra-frame and inter-frame iterative strategy of the recurrent neural network; inputting the preprocessed training data into a recurrent neural network, and guiding a training process of networkparameters by using linear weighting of four losses to obtain a trained model; and inputting the low-quality jitter test video into the trained neural network to obtain a stable version of the targetvideo. According to the invention, the historical motion state information is transmitted to each current video frame through the iterative strategy of the recurrent neural network in the aspect of time sequence, and the perception capability of the network for the jitter sequence frame information is enhanced, so that a stable picture can be predicted more accurately.

Description

technical field [0001] The invention belongs to the technical field of remote sensing image processing, in particular to a video stabilization method based on an iterative strategy of a cyclic neural network. Background technique [0002] Remote sensing hyperspectral image super-resolution is a widely used and popular research field. Video is a time-series combination expression of images. Many video processing algorithms are not robust to some low-quality videos (blurry, noise, picture jitter, insufficient light), so video quality is the key to testing the performance of video processing algorithms. . The video image stabilization can be used as a preprocessing step of these algorithms to further improve the algorithm performance by improving the video picture quality. The stabilized video can be better applied to various visual tasks such as super-resolution and classification. [0003] The traditional mainstream video image stabilization method is an image processing m...

AI Technical Summary

Problems solved by technology

However, these deep video stabilization methods only stack adjacent temporal video frames in the input channel dimension, and then design a temporal regularization term to allow the convolutional network to learn the coherence of motion between frames, but this method cannot be well utilized. Timing information of adjacent frames

Images