Creating training data variability in machine learning for object labelling from images

1. Image labeling system ( 100 ) comprising:

a support ( 2 ) for an object ( 3 ) to be labeled;

a digital camera ( 1 ) configured to capture a plurality of images of a scene including said object ( 3 );

a process and control apparatus ( 5 ) configured to receive said plurality of images and generate corresponding labeling data ( 21 - 24 , L 1 -L 4 ) associated to said object ( 3 );

a digital display ( 4 ) associated with said support ( 2 ) and connected to the process and control apparatus ( 5 ) to selectively display additional images ( 7 - 13 ) selected from the group comprising:

first images ( 7 - 11 ) in the form of backgrounds for the plurality of images and introducing a degree of variability in the scene to be captured by the digital camera ( 1 ) and processed by the process and control apparatus ( 5 );

second images ( 12 ) indicating position and/or orientation according to which place said object ( 3 ) by a user on the support ( 2 ) in order to be captured by the digital camera ( 1 );

third images ( 13 ) to be captured by the digital camera ( 1 ) and provided to the process and control apparatus ( 5 ) to evaluate a position of the digital camera ( 1 ) with respect the digital display ( 4 );

fourth images to be produced when the object ( 3 ) is already placed on the support ( 2 ), captured by the digital camera ( 1 ) and provided to the process and control apparatus ( 5 ) to evaluate at least one of the following data of the object ( 3 ): position, orientation, 3D shape.

2. System ( 100 ) according to claim 1 , wherein the process and control apparatus ( 5 ) is configured to cause the digital display ( 4 ) to selectively display the first images ( 7 - 11 ) comprising a first background image ( 7 ) having corresponding colour and texture and a second background image ( 10 ) having relative colour and/or texture different from said colour and texture of the first background image.

3. System ( 100 ) according to claim 1 , wherein:

said digital camera ( 1 ) is movable with respect to said support ( 2 );

the process and control apparatus ( 5 ) is configured to store the following data: optical parameters of said digital camera ( 1 ), position data and orientation data of said digital camera ( 1 ), position data and orientation data of said object ( 3 ).

4. System ( 100 ) according to claim 1 , wherein said corresponding labeling data are selected from the following data group: text description of the scene represented in the captured image, 2D Bounding Contour ( 21 , 22 , 24 ), 3D Bounding Shape, 3D Pose information ( 23 ), Depth Map, Semantic Map (L 1 -L 4 ).

5. System according to claim 2 , further provided with a movable support ( 6 ) for supporting and displacing the digital camera ( 1 ) or the digital display ( 4 ) to capture said scene according to different point of views.

6. System according to claim 5 , wherein the movable support is a mechanical manipulator ( 6 ) selected from the group: cartesian/gantry robot, cylindrical robot, spherical robot, SCARA robot, articulated robot, parallel robot.

7. System according to claim 5 , wherein the movable support is a passive arm.

8. System according to claim 2 , wherein digital camera ( 1 ) is hand supported and comprises a plurality of sensors to provide said position and orientation data of the digital camera ( 1 ) to the process and control apparatus ( 5 ).

9. System according to claim 1 , wherein the process and control apparatus ( 5 ) is configured to display the third images ( 13 ) with a pre-established frequency to avoid that the third images ( 13 ) appear in said plurality of images of a scene including said object ( 3 ).

10. System according to claim 6 , wherein the mechanical manipulator ( 6 ) further comprises a gripper ( 14 ) configured to grab the object ( 3 ).