RECOGNITION MODEL TRAINING METHOD AND APPARATUS, AND MOBILE INTELLIGENT DEVICE
This application relates to the field of intelligent driving technologies. A recognition model training method includes: obtaining a first image captured by a first photographing component, where the first photographing component corresponds to a first angle of view; generating, based on the first image, a second image corresponding to a second photographing component, where the second photographing component is determined based on the first photographing component, and the second photographing component corresponds to a second angle of view; determining, based on first three-dimensional information of a first object in the first image, second three-dimensional information of the first object in the second image; and training a recognition model based on the second image and the second three-dimensional information of the first object, where the recognition model is used to recognize an object in an image captured by the first photographing component.
1 . A recognition model training method, wherein the method comprises:
obtaining a first image captured by a first photographing component, wherein the first photographing component corresponds to a first angle of view;
generating, based on the first image, a second image corresponding to a second photographing component, wherein the second photographing component is determined based on the first photographing component, and the second photographing component corresponds to a second angle of view;
determining, based on first three-dimensional information of a first object in the first image, second three-dimensional information of the first object in the second image; and
training a recognition model based on the second image and the second three-dimensional information, wherein the recognition model is used to recognize an object in an image captured by the first photographing component.
2 . The method according to claim 1 , wherein the second photographing component is a virtual photographing component, and before the generating, based on the first image, a second image corresponding to a second photographing component, the method further comprises:
rotating a coordinate system of the first photographing component by a preset angle to obtain the second photographing component.
3 . The method according to claim 1 , wherein
intrinsic parameters of the first photographing component and the second photographing component are the same; or
intrinsic parameters of the first photographing component and the second photographing component are different.
4 . The method according to claim 1 , wherein the generating, based on the first image, a second image corresponding to a second photographing component comprises:
determining first coordinates that are of a first pixel in the first image and that are on a preset reference surface, wherein the first pixel is any pixel in the first image;
determining, based on the first coordinates, second coordinates that are of the first pixel and that correspond to the second photographing component;
determining a second pixel based on the second coordinates; and
generating the second image based on the second pixel.
5 . The method according to claim 4 , wherein the preset reference surface is a spherical surface that uses an optical center of the first photographing component as a sphere center.
6 . The method according to claim 1 , wherein the determining, based on first three-dimensional information of a first object in the first image, second three-dimensional information of the first object in the second image comprises:
determining the second three-dimensional information based on the first three-dimensional information and a coordinate conversion relationship between the first photographing component and the second photographing component.
7 . The method according to claim 1 , wherein before the determining, based on first three-dimensional information of a first object in the first image, second three-dimensional information of the first object in the second image, the method further comprises:
obtaining point cloud data that corresponds to the first image and that is collected by a sensor, wherein the point cloud data comprises third three-dimensional information of the first object; and
determining the first three-dimensional information based on the third three-dimensional information and a coordinate conversion relationship between the first photographing component and the sensor.
8 . A recognition model training apparatus, comprising:
at least one processor; and
one or more memories coupled to the at least one processor and storing programming instructions for execution by the at least one processor to cause the apparatus to:
obtain a first image captured by a first photographing component, wherein the first photographing component corresponds to a first angle of view;
generate, based on the first image, a second image corresponding to a second photographing component, wherein the second photographing component is determined based on the first photographing component, and the second photographing component corresponds to a second angle of view;
determine, based on first three-dimensional information of a first object in the first image, second three-dimensional information of the first object in the second image; and
train a recognition model based on the second image and the second three-dimensional information, wherein the recognition model is used to recognize an object in an image captured by the first photographing component.
9 . The apparatus according to claim 8 , wherein the second photographing component is a virtual photographing component; and the programming instructions, when executed by the at least one processor, cause the apparatus to: rotate a coordinate system of the first photographing component by a preset angle to obtain the second photographing component.
10 . The apparatus according to claim 8 , wherein
intrinsic parameters of the first photographing component and the second photographing component are the same; or
intrinsic parameters of the first photographing component and the second photographing component are different.
11 . The apparatus according to claim 8 , wherein the programming instructions, when executed by the at least one processor, cause the apparatus to:
determine first coordinates that are of a first pixel in the first image and that are on a preset reference surface, wherein the first pixel is any pixel in the first image;
determine, based on the first coordinates, second coordinates that are of the first pixel and that correspond to the second photographing component;
determine a second pixel based on the second coordinates; and
generate the second image based on the second pixel.
12 . The apparatus according to claim 11 , wherein the preset reference surface is a spherical surface that uses an optical center of the first photographing component as a sphere center.
13 . The apparatus according to claim 8 , wherein the programming instructions, when executed by the at least one processor, cause the apparatus to:
determine the second three-dimensional information based on the first three-dimensional information and a coordinate conversion relationship between the first photographing component and the second photographing component.
14 . The apparatus according to claim 8 , wherein the programming instructions, when executed by the at least one processor, cause the apparatus to:
obtain point cloud data that corresponds to the first image and that is collected by a sensor, wherein the point cloud data comprises third three-dimensional information of the first object; and
determine the first three-dimensional information based on the third three-dimensional information and a coordinate conversion relationship between the first photographing component and the sensor.
15 . A mobile intelligent device, comprising a first photographing component and a recognition model training apparatus, wherein the recognition model training apparatus comprises:
at least one processor; and
one or more memories coupled to the at least one processor and storing programming instructions for execution by the at least one processor to cause the recognition model training apparatus to:
receive a first image from the first photographing component, wherein the first photographing component corresponds to a first angle of view;
generate, based on the first image, a second image corresponding to a second photographing component, wherein the second photographing component is determined based on the first photographing component, and the second photographing component corresponds to a second angle of view;
determine, based on first three-dimensional information of a first object in the first image, second three-dimensional information of the first object in the second image; and
train a recognition model based on the second image and the second three-dimensional information, wherein the recognition model is used to recognize an object in an image captured by the first photographing component; and
wherein the first photographing component is configured to: capture the first image, and transmit the first image to the recognition model training apparatus.
16 . The mobile intelligent device according to claim 15 , wherein the second photographing component is a virtual photographing component; and the programming instructions, when executed by the at least one processor, cause the recognition model training apparatus to: rotate a coordinate system of the first photographing component by a preset angle to obtain the second photographing component.
17 . The mobile intelligent device according to claim 15 , wherein
intrinsic parameters of the first photographing component and the second photographing component are the same; or
intrinsic parameters of the first photographing component and the second photographing component are different.
18 . The mobile intelligent device according to claim 15 , wherein the programming instructions, when executed by the at least one processor, cause the recognition model training apparatus to:
determine first coordinates that are of a first pixel in the first image and that are on a preset reference surface, wherein the first pixel is any pixel in the first image;
determine, based on the first coordinates, second coordinates that are of the first pixel and that correspond to the second photographing component;
determine a second pixel based on the second coordinates; and
generate the second image based on the second pixel.
19 . The mobile intelligent device according to claim 19 , wherein the preset reference surface is a spherical surface that uses an optical center of the first photographing component as a sphere center.