Object recognition system, recognition device, object recognition method, and object recognition program

1. An object recognition system comprising:

a recognition device that recognizes an object in a captured image; and

a server that generates a learning model,

wherein the recognition device includes:

a first object recognition unit that determines a type of the object in the image using the learning model; and

an image transmission unit that transmits a type-indeterminable image, which is an image in which the type has not been determined, to the server when the object included in the type-indeterminable image is an object detected as a three-dimensional object,

the server includes:

a learning device that generates a learning model based on training data in which a teacher label is assigned to the type-indeterminable image; and

a learning model transmission unit that transmits the generated learning model to the recognition device, and

the first object recognition unit of the recognition device determines the type of the object in the image using the transmitted learning model.

2. The object recognition system according to claim 1 , wherein

the server includes a second object recognition unit that determines the type of the object in the type-indeterminable image received from the recognition device using a learning model with higher recognition accuracy than the learning model used by the first object recognition unit.

3. The object recognition system according to claim 1 , wherein

the recognition device includes a position information transmission unit that transmits position information indicating a location where a type-indeterminable image has been captured to the server,

the server includes:

a position collection unit that collects the received position information; and

a priority determination unit that determines a priority of the type-indeterminable image captured at the location indicated by the position information to be higher as the collected position information is more, and

the learning device generates a learning model preferentially using training data including a type-indeterminable image with a higher priority.

4. The object recognition system according to claim 3 , wherein

the server includes a transmission permission position information transmission unit that transmits transmission permission position information, which is position information indicating a capturing location of a type-indeterminable image for which transmission to the server is permitted, to the recognition device,

the position collection unit collects pieces of position information in which positions are close to each other in a group, and identifies a dense location which is a location where type-indeterminable images have been captured more than a predetermined standard,

the transmission permission position information transmission unit transmits information indicating the identified dense location to the recognition device as the transmission permission position information, and

the image transmission unit of the recognition device suppresses transmission of a type-indeterminable image captured at a location other than the location indicated by the transmission permission position information to the server.

5. The object recognition system according to claim 3 , wherein

the recognition device includes a time information transmission unit that transmits time information indicating time when a type-indeterminable image has been captured to the server, and

the priority determination unit determines a priority to be set for the type-indeterminable image captured at the dense location according to a priority policy which is a policy that defines a method for determining a priority of an image used for learning, and is a policy that is defined based on a duration of time when the image has been captured.

6. The object recognition system according to claim 1 , wherein

the recognition device includes:

a position information transmission unit that transmits position information on a location where a type-indeterminable image has been captured to the server; and

an orientation information transmission unit that transmits orientation information indicating an orientation in which the type-indeterminable image has been captured to the server,

the server includes an image data acquisition unit that acquires an image identified using the received position information and orientation information from an image database capable of identifying an image based on the position information and the orientation information, and

the learning device generates a learning model based on training data in which a teacher label is assigned to the image acquired by the image data acquisition unit.

7. The object recognition system according to claim 6 , wherein

the recognition device includes a camera information transmission unit that transmits a camera parameter related to an imaging device that has captured the type-indeterminable image to the server, and

the image data acquisition unit corrects the image acquired from the image database based on the received camera parameter.

8. The object recognition system according to claim 1 , wherein

the image transmission unit transmits an image in which a captured image is associated with coordinate information for identifying an object whose type has not been determined in the captured image, or an image obtained by extracting a portion of the captured image including the object whose type has not been determined as a type-indeterminable image.

9. A recognition device comprising a hardware processor configured to execute a software code to:

determine a type of an object in an image using a learning model;

transmit a type-indeterminable image, which is an image in which the type has not been determined, to a server that generates the learning model when an object included in the type-indeterminable image is an object detected as a three-dimensional object; and

receive the learning model generated by the server based on training data in which a teacher label is assigned to the type-indeterminable image, and determine the type of the object in the image using the received learning model.

10. An object recognition method comprising:

causing a recognition device that recognizes an object in a captured image to determine a type of the object in the image using a learning model;

causing the recognition device to transmit a type-indeterminable image, which is an image in which the type has not been determined, to a server that generates the learning model when an object included in the type-indeterminable image is an object detected as a three-dimensional object;

causing the server to generate a learning model based on training data in which a teacher label is assigned to the type-indeterminable image;

causing the server to transmit the generated learning model to the recognition device; and

causing the recognition device to determine the type of the object in the image using the transmitted learning model.