Method for generating three-dimensional point cloud by single image based on CNN (Convolutional Neural Network)

Abstract

The invention provides a CNN-based method for generating a three-dimensional point cloud through a single image, and the method is characterized in that the method comprises the following steps: S1, collecting a plurality of single images, and taking the collected single images as a training data set; and S2, constructing an image encoder. And S3, inputting the training data set into an image encoder. And S4, constructing a point cloud automatic encoder. And step S5, training the point cloud automatic encoder by using a ChamferDistance loss function. And S6, constructing an image three-dimensional point cloud reconstruction network model. And step S7, carrying out training by using an EarthMover's Distance loss function. And S8, inputting a single image to be detected into the trained image three-dimensional point cloud reconstruction network model to obtain the three-dimensional point cloud of the single image to be detected. According to the method, the attention mechanism is added in the reconstruction network, and the single image is firstly generated into the simple point cloud and then generated into the accurate point cloud, so that the precision of three-dimensional reconstruction of the single image is improved, and the problem of low reconstruction precision caused by direct output of the point cloud in three-dimensional reconstruction of the single image is solved.

Description

technical field [0001] The invention relates to a method for generating a three-dimensional point cloud from a CNN-based single image. Background technique [0002] Computer vision is a camera that takes images, and uses computers to identify and detect objects in the images. It can be said that computer vision is a science that studies how to make machines "see". It collects pictures or videos, processes and analyzes the pictures or videos, and obtains useful information from them. Image-based 3D reconstruction technology is one of the key research areas of computer vision, that is, using technology to process 2D images to reconstruct a 3D target model. There are a wide range of applications. [0003] Image-based 3D reconstruction can be divided into monocular vision methods and multi-eye vision methods according to image acquisition methods. The 3D reconstruction based on multi-eye vision needs to preprocess the image and find the feature points for matching between the...

AI Technical Summary

Problems solved by technology

In the image processing network, as the network level increases, not only will it consume a lot of computing power of the device, but the training effect may also be unsatisfactory.

Images