Video multi-frame reconstruction method and device based on single-pixel camera

Abstract

A video multi-frame reconstruction method and device based on a single-pixel camera, the method uses a single-pixel camera to obtain one-dimensional measurement values ​​corresponding to continuous multi-frame images; establishes a video multi-frame reconstruction model based on the one-dimensional measurement values, and the video multi-frame reconstruction model Make the multi-frame image have low rank in the time dimension, and have a full variation regularization prior in the space dimension; solve the video multi-frame reconstruction model by the augmented Lagrangian multiplier method, and obtain each variable iteration Expression; Substituting the measured value and the known modulation mask matrix, after multiple iterations, multiple frames of images are decoupled from the measured value. Therefore, the embodiment of the present application under the constraints of low-rank priors between multiple frames of images and full variational regularization within video frames can reconstruct multiple frames of images at the same time at a lower sampling rate, improving the reconstruction efficiency and reconstruction efficiency. Accuracy, robustness to noise, that is to say, the embodiment of the present application has characteristics such as high signal-to-noise ratio, high precision, low cost, high efficiency, and good adaptability.

Description

technical field [0001] The invention relates to the technical fields of computer vision and computational photography, in particular to a single-pixel camera-based video multi-frame reconstruction method and device. Background technique [0002] A single-pixel camera is a multi-dimensional imaging device that utilizes a spatial light modulator and a single photosensitive detector. The single-pixel imaging method breaks the limitation that photosensitive detectors can only achieve single-point detection, and has the advantages of low cost, high signal-to-noise ratio, and wide spectral range. Therefore, single-pixel cameras have been widely used in hyperspectral imaging, terahertz imaging, 3D imaging, and optical encryption. [0003] In recent years, the superior performance of single-pixel cameras and the demand for video imaging have made single-pixel video reconstruction a hot field. Single-pixel video imaging speed is limited by two factors: acquisition speed and reconst...

AI Technical Summary

Problems solved by technology

The method based on the Walsh-Hamdard transform is the mainstream of the existing single-pixel video traditional methods. The reconstruction speed of the Walsh-Hamdard transform is fast and the sampling rate is low, but the efficiency still needs to be improved.

In 2018, single-pixel video imaging based on convolutional auto-encoding network reconstructed 128*128 images at 30 frames per second and 0.02 sampling rate, with high precision and efficiency, but depended on training samples and poor repeatability

Images