Processing device for conversion of images
A processing unit recognizes an object based on image data. An object recognition unit identifies the object based on the image data. A conversion unit is configured as a neural network provided as an upstream stage of the object recognition unit. The conversion unit converts a first image IMG acquired by the camera into a second image, and inputs the second image to the object recognition unit.
1. A processing device structured to recognize an object, comprising:
a conversion unit structured as a neural network, and structured to receive, from a camera, a two-dimensional first image captured by the camera in a first environment and convert the two-dimensional first image into a two-dimensional second image associated with a second environment different from the first environment; and
an object recognition unit structured to identify an object based on the two-dimensional second image, wherein
the object recognition unit is trained using learning image data captured under the second environment,
and wherein the two-dimensional second image is an image to be acquired if the same scene was captured in the second environment.
2. The processing device according to claim 1 , wherein the two-dimensional second image is obtained by correcting shades of the two-dimensional first image such that they are an approximation of learning data used for training of the object recognition unit.
3. The processing device according to claim 1 , wherein the two-dimensional second image of the same scene as that of the two-dimensional first image is generated as an image as acquired in the same environment as an environment in which the learning data used for the training of the object recognition unit was acquired.
4. The processing device according to claim 3 , wherein the learning data is acquired in daytime,
and wherein the conversion unit converts a first image acquired in nighttime into a second image as acquired in daytime.
5. The processing device according to claim 1 , wherein, in the training of the conversion unit, the neural network of the conversion unit is optimized with reference to an identification rate of the object recognition unit so as to improve the identification rate.
6. The processing device according to claim 1 , wherein the conversion unit receives a plurality of consecutive frames as input.
7. An object identification system comprising:
a camera; and
the processing device according to claim 1 .
8. An automotive lamp comprising the object identification system according to claim 7 .
9. An automobile comprising:
a camera built into a headlamp; and
the processing device according to claim 1 .
10. A processing device, comprising:
a conversion unit structured to receive, from a camera, a two-dimensional first image captured by the camera in a first environment and convert the two-dimensional first image into a two-dimensional second image associated with a second environment different from the first environment; and
an object recognition unit structured to process the two-dimensional second image processed by the conversion unit, and to identify the object,
and wherein the object recognition unit is trained using a two-dimensional image captured in the second environment as learning data; and
the conversion unit is trained using a set of a two-dimensional image captured in the second environment and a two-dimensional image captured in the first environment that differs from the second environment,
wherein the two-dimensional second image is an image to be acquired if the same scene was captured in the second environment.