A Depth Prediction Method of Monocular Image Based on Neural Network

Abstract

The invention discloses a neural network-based monocular image depth prediction method, which constructs a neural network including an input layer, a hidden layer and an output layer, the hidden layer includes an encoding and decoding network framework, and the encoding network framework includes five neural network blocks , 1 reasoning layer and 1 connection layer, the first and second neural network blocks are composed of 2 convolutional layers and 1 maximum pooling layer, and the 3rd to 5th neural network blocks are composed of 3 Consisting of three convolutional layers and one maximum pooling layer, the reasoning layer includes two perforated convolutional neural networks, and the decoding network framework includes five neural network blocks, five connection layers, and four independent bilinear upsampling layers. Each neural network block consists of 1 convolutional layer and 1 bilinear upsampling layer; the monocular images in the training set are input into the neural network for training; the predicted monocular images are input into the neural network model during testing Prediction is carried out in the prediction depth image; the advantage is that the prediction accuracy is high and the calculation complexity is low.

Description

technical field [0001] The invention relates to an image depth prediction technology, in particular to a neural network-based monocular image depth prediction method. Background technique [0002] With the rapid development of machine learning, it has become possible for machines to imitate humans to estimate the distance of objects from images to a certain extent, that is, machines can predict the depth of a single image and obtain a depth map to a certain extent. Depth maps are widely used in 3D reconstruction, robot navigation and other fields. At the same time, depth maps are helpful in computer vision fields such as detection and segmentation because they provide information about the distance of objects. Depth maps are currently mainly derived from different types of depth cameras. However, depth cameras have limitations in application due to their shortcomings such as high price and low portability. However, the cost of depth information extracted by monocular vision ...

AI Technical Summary

Problems solved by technology

However, the above methods do not fully consider the global and local features, so there is still room for improvement in the accuracy of depth prediction

Images