FIELD: image processing.
SUBSTANCE: invention relates to the field of image processing, and, in particular, it is oriented to the building of a 3D-model of an object of images obtained in conditions of limited availability of the object for observance and measurement of its parameters. A method for the reconstruction of 3D-model of an object is claimed, according to which an available set of full-scale images of the object is obtained, on which training of a convolutional neural network is performed. Firstly, a set of current versions of the 3D-model of the object is created based on the obtained set of full-scale images of the object, image sets are formed of each current version of the 3D-model of the object, the object is detected on image sets, using the trained convolutional neural network. Current values of the probability of detection of the object on image sets are calculated, M>2 the most values of the probability of detection of the object and M current versions of the 3D-model of the object, corresponding to them, are selected among calculated current values of the probability of detection of the object. A set of current versions of the 3D-model of the object is created from each selected version of the 3D-model of the object by changing at least one parameter of its shape, geometrical dimensions, color textures, and surface reflectivity. Image sets are repeatedly formed of each current version of the 3D-model of the object, and subsequent actions are performed until value of at least one of the most values of the probability of detection of the object is increased, otherwise the current version of the 3D-model of the object with the most value of the probability of detection of the object is taken as the reconstructed 3D-model of the object.
EFFECT: increase in the accuracy of building of a model in conditions of limited availability of an object for observance and measurement of its parameters.
4 cl, 9 dwg
| Title | Year | Author | Number |
|---|---|---|---|
| IMAGE PROCESSING METHOD WITH TRAINED NEURAL NETWORKS | 2021 |
|
RU2779281C1 |
| METHOD OF OBTAINING INFORMATION ON SHAPE AND DIMENSIONS OF THREE-DIMENSIONAL OBJECT FROM ITS TWO-DIMENSIONAL IMAGE | 2022 |
|
RU2816504C1 |
| METHOD FOR PROCESSING IMAGES BY CONVOLUTIONAL NEURAL NETWORKS | 2020 |
|
RU2771442C1 |
| NEURAL DOT GRAPHIC | 2019 |
|
RU2729166C1 |
| VISUALIZATION OF RECONSTRUCTION OF 3D SCENE USING SEMANTIC REGULARIZATION OF NORMALS TSDF WHEN TRAINING NEURAL NETWORK | 2023 |
|
RU2825722C1 |
| METHOD OF RECOGNIZING SPEECH EMOTIONS USING 3D CONVOLUTIONAL NEURAL NETWORK | 2023 |
|
RU2816680C1 |
| IMAGE PROCESSING METHOD, AN IMAGE PROCESSING DEVICE AND A DATA MEDIUM | 2018 |
|
RU2709437C1 |
| NEURAL NETWORK TRANSFER OF THE FACIAL EXPRESSION AND POSITION OF THE HEAD USING HIDDEN POSITION DESCRIPTORS | 2020 |
|
RU2755396C1 |
| METHOD FOR PROVIDING COMPUTER VISION | 2022 |
|
RU2791587C1 |
| METHOD AND ELECTRONIC DEVICE FOR DETECTING THREE-DIMENSIONAL OBJECTS USING NEURAL NETWORKS | 2021 |
|
RU2776814C1 |
