FIELD: imaging.
SUBSTANCE: invention relates to tools for creating two-dimensional images of a three-dimensional scene visible from different view points. The technical result is to synthesize from dot clouds a two-dimensional image of the scene viewed from the desired viewing point, with high quality and low computational costs. A three-dimensional cloud of dots is received, obtained from multiple two-dimensional images of the same scene, wherein each dot of the cloud is defined by three-dimensional coordinates in the world coordinate system and by a vector representation of the dot. A view point is set as a camera with internal parameters and external parameters. The three-dimensional coordinates of each dot are converted into two-dimensional coordinates and the depth of each dot in the coordinate system of the screen space of the camera, using the internal parameters and external parameters; a plurality of beams diverging from the point of view is set, wherein the beams are defined by the coordinates of the screen space and the internal parameters and external parameters. Dots are grouped into sets of dots associated with beams, wherein each set of dots comprises dots wherein one beam passes though said dots, and in each set of dots, the dots are arranged in order of decreasing depth thereof relative to the view point. The vector representation of the beam is calculated for each beam by aggregating the vector representations of the dots and the depths of the corresponding set of dots using a trained machine learning predictor. The vector representations of beams are projected onto the image plane, wherein the previous stages are executed for the set plurality of scales. Image planes are merged within the plurality of scales using a trained machine learning predictor into a two-dimensional image.
EFFECT: technical result is synthesis of a two-dimensional image with high quality and low computational costs of a scene viewed from the required point of view from clouds of dots.
6 cl, 5 dwg
Title | Year | Author | Number |
---|---|---|---|
TEXTURED NEURAL AVATARS | 2019 |
|
RU2713695C1 |
NEURAL DOT GRAPHIC | 2019 |
|
RU2729166C1 |
METHOD FOR PROVIDING COMPUTER VISION | 2022 |
|
RU2791587C1 |
METHOD AND ELECTRONIC DEVICE FOR DETECTING THREE-DIMENSIONAL OBJECTS USING NEURAL NETWORKS | 2021 |
|
RU2776814C1 |
METHOD FOR VISUALIZING A 3D PORTRAIT OF A PERSON WITH ALTERED LIGHTING AND A COMPUTING DEVICE FOR IT | 2021 |
|
RU2757563C1 |
METHOD FOR CONSTRUCTING A SCENE REPRESENTATION WITH DIRECT CORRECTION FOR REAL-TIME IMAGE SYNTHESIS | 2022 |
|
RU2799237C1 |
METHOD OF FORMATION OF NEURAL NETWORK ARCHITECTURE FOR CLASSIFICATION OF OBJECT TAKEN IN CLOUD OF POINTS, METHOD OF ITS APPLICATION FOR TEACHING NEURAL NETWORK AND SEARCHING SEMANTICALLY ALIKE CLOUDS OF POINTS | 2017 |
|
RU2674326C2 |
METHODS AND SYSTEMS FOR DETERMINING THE PRESENCE OF DYNAMIC OBJECTS BY A COMPUTER | 2019 |
|
RU2767955C1 |
VISUALIZATION OF RECONSTRUCTION OF 3D SCENE USING SEMANTIC REGULARIZATION OF NORMALS TSDF WHEN TRAINING NEURAL NETWORK | 2023 |
|
RU2825722C1 |
MODELLING PEOPLE'S CLOTHING BASED ON MULTIPLE POINTS | 2021 |
|
RU2776825C1 |
Authors
Dates
2021-06-16—Published
2020-04-15—Filed