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

1. An object recognition device comprising:

a camera configured to capture an image of a shooting location where an object is to be placed; and

a processor configured to recognize an object included in an image of the shooting location, utilizing a machine learning model for object recognition;

wherein the machine learning model for object recognition is constructed based on trained model data generated by a training operation using a composite image of a backgroundless object image and a background image for training acquired by capturing an image of the shooting location where no object is placed, and

wherein the processor is configured to:

determine necessity of an update operation on the machine learning model for object recognition at a predetermined time;

when determining that the update operation is necessary, cause the camera to capture an image of the shooting location where no object is placed to thereby re-acquire a background image for training; and

cause the machine learning model to be trained using a composite image of a backgroundless object image and the re-acquired background image for training as training data,

wherein the camera captures a current image of the shooting location where no object is placed, to thereby acquire a background image for determination, and

wherein the processor determines that the update operation is necessary when the background image for determination is different from the background image for training to an amount equal to or greater than a predetermined level.

2. The object recognition device according to claim 1 , wherein machine learning models for object recognition are created for a plurality of times of day, and

wherein the processor determines which of the machine learning models for object recognition needs to be used in the update operation based on the time of day when the background image for determination is different from the background image for training to an amount equal to or greater than the predetermined level.

3. The object recognition device according to claim 1 , wherein machine learning models for object recognition are created for a plurality of types of weather conditions, and

wherein the processor determines which of the machine learning models for object recognition needs to be used in the update operation based on the weather condition at a time when the background image for determination is different from the background image for training to an amount equal to or greater than the predetermined level.

4. The object recognition device according to claim 1 , wherein the processor determines that the update operation is necessary when at least one of a current installation position and a current orientation of the object recognition device is different from that at the time of the training operation, to an amount equal to or greater than a predetermined level.

5. The object recognition device according to claim 1 , wherein the processor determines that the update operation is necessary based on a number of times which inconsistency is detected between a result of detection of a user of the object recognition device and a result of detection of an object placed at the shooting location.

6. The object recognition device according to claim 5 , wherein the processor determines that there is inconsistency between a result of detection of the user and a result of detection of the object when the user is not detected and the object is detected.

7. The object recognition device according to claim 5 , wherein the processor determines that there is inconsistency between a result of detection of the user and a result of detection of the object when the user is detected and the object is not detected.

8. The object recognition device according to claim 1 , wherein the processor determines that the update operation is necessary based on a number of times which a user operates to correct an error in a result of an object recognition operation.

9. The object recognition device according to claim 1 , wherein the trained model data is generated by a learning device which holds the backgroundless object image,

wherein the object recognition device further comprises a communication device configured to transmit the background image for training to the learning device and receive the trained model data from the learning device, and

wherein, when the update operation is necessary, the processor causes the communication device to transmit the background image for training to the learning device, thereby causing the learning device to re-perform the training operation.

10. The object recognition device according to claim 1 , wherein the object recognition device is a checkout device for checking out an object placed at the shooting location.

11. An object recognition system comprising a shooting location where an object is to be placed, and a camera for capturing an image of the shooting location and configured to recognize an object included in an image of the shooting location, utilizing a machine learning model for object recognition,

wherein the machine learning model for object recognition is constructed based on trained model data generated by a training operation using a composite image of a backgroundless object image and a background image for training acquired by capturing an image of the shooting location where no object is placed, and

wherein the object recognition system is configured to:

determine necessity of an update operation on the machine learning model for object recognition at a predetermined time;

when determining that the update operation is necessary, cause the camera to capture an image of the shooting location where no object is placed to thereby re-acquire a background image for training; and

cause the machine learning model to be trained using a composite image of a backgroundless object image and the re-acquired background image for training as training data,

wherein the camera captures a current image of the shooting location where no object is placed, to thereby acquire a background image for determination, and

wherein the object recognition system determines that the update operation is necessary when the background image for determination is different from the background image for training to an amount equal to or greater than a predetermined level.

12. An object recognition method comprising:

capturing an image of a shooting location where an object is to be placed with a camera; and

recognizing an object included in an image of the shooting location, utilizing a machine learning model for object recognition,

wherein the object recognition method further comprises:

determining necessity of an update operation on the machine learning model for object recognition at a predetermined time;

when the update operation is necessary, causing the camera to capture an image of the shooting location where no object is placed to thereby re-acquire a background image for training; and

causing the machine learning model to be trained using a composite image of a backgroundless object image and the re-acquired background image for training as training data,

wherein the camera captures a current image of the shooting location where no object is placed, to thereby acquire a background image for determination, and

wherein the update operation is determined to be necessary when the background image for determination is different from the background image for training to an amount equal to or greater than a predetermined level.