Recognition processing device, recognition processing method, and program
Provided are a recognition processing device, a recognition processing method, and a program capable of efficiently narrowing down a three-dimensional region on which recognition processing using a three-dimensional convolutional neural network is to be executed. A first recognition process executing section executes a first recognition process on a captured image obtained by capturing an image of a real space and used to generate voxel data. A target two-dimensional region determining section determines a two-dimensional region occupying part of the captured image on the basis of a result of the first recognition process. A target three-dimensional region determining section determines a three-dimensional region in the real space on the basis of the two-dimensional region and a position of a camera when the camera obtains the captured image. A second recognition process executing section executes a second recognition process using a three-dimensional convolutional neural network on the voxel data associated with a position in the three-dimensional region.
1. A recognition processing device executing recognition processing using a three-dimensional convolutional neural network on voxel data in which a position in a real space and a voxel value are associated with each other, the recognition processing device comprising:
a first recognition process executing section executing a first recognition process on a captured image which is obtained by capturing an image of the real space and which is used to generate the voxel data;
a two-dimensional region determining section determining a two-dimensional region occupying part of the captured image on a basis of a result of the first recognition process;
a three-dimensional region determining section determining a three-dimensional region in the real space on a basis of the two-dimensional region and a position of a camera when the camera obtains the captured image; and
a second recognition process executing section executing a second recognition process using a three-dimensional convolutional neural network on the voxel data associated with a position in the three-dimensional region, wherein:
the first recognition process executing section executes the first recognition process on each of a first captured image and a second captured image obtained by capturing images of the real space from positions different from each other, and
the two-dimensional region determining section determines a first two-dimensional region occupying part of the first captured image and a second two-dimensional region occupying part of the second captured image on a basis of a result of the first recognition process.
2. The recognition processing device according to claim 1 , wherein:
the three-dimensional region determining section determines a first three-dimensional region determined on a basis of the first two-dimensional region and a position of a camera when the camera obtains the first captured image, and a second three-dimensional region determined on a basis of the second two-dimensional region and a position of a camera when the camera obtains the second captured image, and
the second recognition process executing section executes the second recognition process using the three-dimensional convolutional neural network on the voxel data associated with a position in the three-dimensional region in the real space according to the first three-dimensional region and the second three-dimensional region.
3. The recognition processing device according to claim 1 , wherein
the first recognition process executing section executes the first recognition process on a captured image associated with depth information, and
the three-dimensional region determining section determines the three-dimensional region in the real space on a basis of the depth information associated with a position in the two-dimensional region.
4. The recognition processing device according to claim 1 , wherein the first recognition process executing section executes the first recognition process using a two-dimensional convolutional neural network on the captured image.
5. The recognition processing device according to claim 2 , wherein the second recognition process executing section executes the second recognition process using the three-dimensional convolutional neural network on the voxel data associated with a position in the three-dimensional region where the first three-dimensional region and the second three-dimensional region intersect.
6. A recognition processing method for executing recognition processing using a three-dimensional convolutional neural network on voxel data in which a position in a real space and a voxel value are associated with each other, the recognition processing method comprising:
executing a first recognition process on a captured image which is obtained by capturing an image of the real space and which is used to generate the voxel data;
determining a two-dimensional region occupying part of the captured image on a basis of a result of the first recognition process;
determining a three-dimensional region in the real space on a basis of the two-dimensional region and a position of a camera when the camera obtains the captured image; and
executing a second recognition process using a three-dimensional convolutional neural network on the voxel data associated with a position in the three-dimensional region, wherein:
the executing includes executing the first recognition process on each of a first captured image and a second captured image obtained by capturing images of the real space from positions different from each other, and
the determining a two-dimensional region includes determining a first two-dimensional region occupying part of the first captured image and a second two-dimensional region occupying part of the second captured image on a basis of a result of the first recognition process.
7. A non-transitory, computer-readable storage medium containing a program, which when executed by a computer, causes the computer to execute recognition processing using a three-dimensional convolutional neural network on voxel data in which a position in a real space and a voxel value are associated with each other by carrying out actions, comprising:
executing a first recognition process on a captured image which is obtained by capturing an image of the real space and which is used to generate the voxel data;
determining a two-dimensional region occupying part of the captured image on a basis of a result of the first recognition process;
determining a three-dimensional region in the real space on a basis of the two-dimensional region and a position of a camera when the camera obtains the captured image; and
executing a second recognition process using a three-dimensional convolutional neural network on the voxel data associated with a position in the three-dimensional region, wherein:
the executing includes executing the first recognition process on each of a first captured image and a second captured image obtained by capturing images of the real space from positions different from each other, and
the determining a two-dimensional region includes determining a first two-dimensional region occupying part of the first captured image and a second two-dimensional region occupying part of the second captured image on a basis of a result of the first recognition process.