Monocular image estimation method based on multi-classification regression model and self-attention mechanism

Abstract

The invention discloses a monocular image estimation method based on a multi-classification regression model and a self-attention mechanism, and the method comprises the steps: firstly inputting an image, carrying out the replacement of convolution in a convolution unit block through an image encoder, and carrying out the replacement through hole convolution; acquiring pixel-level context information according to a self-attention model after encoding an image encoder, firstly processing an input feature map through a single-layer neural network and a ReLU function, and then acquiring global context information of an image by performing global average pooling on the input feature map; entering scene depth soft inference, classifying input image pixels into depth classes, and performing ordered regression on depth values; and obtaining accurate and smooth depth values by using data provided by a probability map, and obtaining depth inference values of pixels at positions. According to the invention, the depth design of the monocular image scene is carried out by using the ordered classification logistic regression model, the self-attention mechanism and the deep neural network, and the grid effect caused by repeatedly using the same cavity convolution kernel is reduced.

Description

technical field [0001] The invention relates to the technical field of visual positioning, in particular to a monocular image estimation method based on a multi-classification regression model and a self-attention mechanism. Background technique [0002] With the rapid development of science and technology, the spatial resolution and quality of images that can be obtained are getting higher and higher. However, the images obtained by ordinary optical cameras still have great limitations in the application of certain fields, such as A certain smartphone equipped with face recognition function released in 2019 only uses a single front optical camera for matching and recognition, resulting in a vulnerability that the phone can be successfully unlocked by using a pre-prepared photo of the owner. This is because when the monocular image reduces the three-dimensional information to the two-dimensional image information, it lacks the depth information of the scene, and the camera c...

AI Technical Summary

Problems solved by technology

However, there is still a certain gap between this value and the relative error below 0.05 of the current binocular image depth estimation algorithm.

Images