3D auto-labeling with structural and physical constraints

1. A method for 3D auto-labeling of objects with predetermined structural and physical constraints, comprising:

identifying initial object-seeds for all frames from a given frame sequence of a scene;

refining each of the initial object-seeds over the 2D/3D data, while complying with the predetermined structural and physical constraints to auto-label 3D object vehicles within the scene by discarding incorrect auto-labels of the initial object-seeds when the initial object-seeds are identified as contradicting a first vehicle shape prior information; and

linking the auto-label 3D object vehicles over time into trajectories while respecting the predetermined structural and physical constraints by adjusting the linking of the 3D object vehicles over time by applying a second vehicle shape prior information regarding road and physical boundaries to the trajectories.

2. The method of claim 1 , further comprising planning a trajectory of an ego vehicle according to linked trajectories of the auto-label 3D object vehicles while respecting road and physical boundaries.

3. The method of claim 1 , in which identifying the initial object-seeds is performed by a vehicle perception module using 2D/3D data.

4. The method of claim 1 , in which refining the initial object-seeds comprises:

accessing a vehicle shape prior information; and

discarding incorrect auto-labels of the initial object-seeds when the initial object-seeds are identified as contradicting the vehicle shape prior information.

5. The method of claim 1 , in which linking the auto-label 3D object vehicles comprises:

accessing a vehicle shape prior information regarding road and physical boundaries; and

adjusting the linking of the 3D object vehicles over time by applying the road and physical boundaries to the trajectories.

6. The method of claim 1 , further comprising planning a trajectory of an ego vehicle according to perception of the scene from video captured by the ego vehicle.

7. The method of claim 1 , further comprising performing three-dimensional object detection of the auto-label 3D vehicle objects within the scene.

8. The method of claim 1 , further comprising performing three-dimensional pose detection of the auto-label 3D vehicle objects within the scene.

9. A non-transitory computer-readable medium having program code recorded thereon for 3D auto-labeling of objects with predetermined structural and physical constraints, the program code being executed by a processor and comprising:

program code to identify initial object-seeds for all frames from a given frame sequence of a scene;

program code to refine each of the initial object-seeds over the 2D/3D data, while complying with the predetermined structural and physical constraints to auto-label 3D object vehicles within the scene by program code to discard incorrect auto-labels of the initial object-seeds when the initial object-seeds are identified as contradicting a first vehicle shape prior information; and

program code to link the auto-label 3D object vehicles over time into trajectories while respecting the predetermined structural and physical constraints by program code to adjust the linking of the 3D object vehicles over time by applying a second vehicle shape prior information regarding road and physical boundaries to the trajectories.

10. The non-transitory computer-readable medium of claim 9 , further comprising program code to plan a trajectory of an ego vehicle according to linked trajectories of the auto-label 3D object vehicles while respecting road and physical boundaries.

11. The non-transitory computer-readable medium of claim 9 , in which the program code to identify the initial object-seeds is performed by a vehicle perception module using 2D/3D data.

12. The non-transitory computer-readable medium of claim 9 , in which the program code to refine the initial object-seeds comprises:

program code to access a vehicle shape prior information; and

program code to discard incorrect auto-labels of the initial object-seeds when the initial object-seeds are identified as contradicting the vehicle shape prior information.

13. The non-transitory computer-readable medium of claim 9 , in which program code to link the auto-label 3D object vehicles comprises:

program code to access a vehicle shape prior information regarding road and physical boundaries; and

program code to adjust the linking of the 3D object vehicles over time by applying the road and physical boundaries to the trajectories.

14. The non-transitory computer-readable medium of claim 9 , further comprising program code to plan a trajectory of an ego vehicle according to perception of the scene from video captured by the ego vehicle.

15. The non-transitory computer-readable medium of claim 9 , further comprising program code to perform three-dimensional object detection of the auto-label 3D vehicle objects within the scene.

16. The non-transitory computer-readable medium of claim 9 , further comprising program code to perform three-dimensional pose detection of the auto-label 3D vehicle objects within the scene.

17. A system for 3D auto-labeling of objects with predetermined structural and physical constraints, the system comprising:

an object-seed detection module trained to identify initial object-seeds for all frames from a given frame sequence of a scene;

an object-seed refinement module trained to refine each of the initial object-seeds over the 2D/3D data, while complying with the predetermined structural and physical constraints to auto-label 3D object vehicles within the scene by discarding incorrect auto-labels of the initial object-seeds when the initial object-seeds are identified as contradicting a first vehicle shape prior information; and

a 3D auto-labeling module trained to link the auto-label 3D object vehicles over time into trajectories while respecting the predetermined structural and physical constraints by adjusting the linking of the 3D object vehicles over time by applying a second vehicle shape prior information regarding road and physical boundaries to the trajectories.

18. The system of claim 17 , further comprising a vehicle trajectory module trained to plan a trajectory of an ego vehicle according to linked trajectories of the auto-label 3D object vehicles while respecting road and physical boundaries.

19. The system of claim 17 , further comprising a vehicle perception module trained to identify the initial object-seeds is performed using 2D/3D data.

20. The system of claim 17 , in which the object-seed refinement module is further trained:

to access a vehicle shape prior information; and

to discard incorrect auto-labels of the initial object-seeds when the initial object-seeds are identified as contradicting the vehicle shape prior information.