Bounding box estimation and lane vehicle association

1. A method comprising:

determining a 2D bounding box (2DBB) associated with a vehicle within a camera frame; and

estimating a 3DBB associated with the vehicle based on the 2DBB and lane information, the 3DBB comprising a first box and a second box, the first box associated with a first portion of the vehicle and the second box associated with a second portion of the vehicle,

wherein the lane information comprises information on road geometry of one or more lanes within a field of view of one or more sensors of an ego vehicle,

wherein the first and second boxes together bound an entirety of the vehicle,

wherein the 2DBB is a truncated BB associated with a partial view of the vehicle, and

wherein estimating the 3DBB comprises:

selecting a pseudo vehicle based on the 2DBB;

generating one or more pseudo 3DBBs associated with the pseudo vehicle on corresponding one or more lanes; and

selecting one of the pseudo 3DBBs based a comparison of the one or more pseudo 3DBBs and the 2DBB.

2. The method of claim 1 , wherein the 2DBB bounds the vehicle.

3. The method of claim 1 , wherein the first and second boxes share a common edge.

4. The method of claim 1 , wherein selecting the pseudo vehicle comprises:

determining a class of the vehicle based on the 2DBB; and

selecting the pseudo vehicle from a plurality of class vehicles based on the determined class of the vehicle.

5. The method of claim 1 , wherein selecting one of the pseudo 3DBBs comprises:

for each pseudo 3DBB, projecting the pseudo 3DBB such that a vertical side of the projected 3DBB is aligned with a vertical side of the 2DBB;

for each pseudo 3DBB, comparing the projected 3DBB with the 2DBB; and

selecting one of the pseudo 3DBBs based on the comparison.

6. The method of claim 1 , wherein generating the one or more pseudo 3DBBs comprises:

determining the one or more lanes within the camera frame; and

generating the one or more 3DBB of the pseudo vehicle placed in the one or more lanes.

7. The method of claim 6 , wherein the one or more lanes are determined from processing prior camera frames and/or from a lane map.

8. The method of claim 6 , wherein the one or more lanes are determined from a lane map.

9. A method comprising:

determining a 2D bounding box (2DBB) associated with a vehicle within a camera frame;

estimating a 3DBB associated with the vehicle based on the 2DBB and lane information, the 3DBB comprising a first box and a second box, the first box associated with a first portion of the vehicle and the second box associated with a second portion of the vehicle; and

updating the estimated 3DBB,

wherein the lane information comprises information on road geometry of one or more lanes within a field of view of one or more sensors of an ego vehicle, and

wherein the first and second boxes together bound an entirety of the vehicle,

wherein updating the estimated 3DBB comprises:

projecting the estimated 3DBB onto the 2DBB;

extracting the first portion of the vehicle from the 2DBB based on the projected estimated 3DBB;

determining lane boundaries of a lane the vehicle is located within the camera frame;

determining a position of the vehicle within the lane of the vehicle;

estimating a 3D depth of the vehicle; and

updating the estimate 3DBB based on the 3D depth.

10. The method of claim 9 , wherein determining the lane boundaries comprises lane curve fitting on 2D lane vertices.

11. The method of claim 10 , wherein the 2D lane vertices are obtained from lane information, the lane information included in a lane map or derived from processing prior camera frames.

12. The method of claim 9 , wherein determining the position of the vehicle within the lane comprises:

determining 2D crossing points between the lane boundaries and a line extended from a bottom side of the first portion of the vehicle; and

determining the position of the vehicle as a position of a center of the vehicle relative to a midpoint of the 2D crossing points.

13. The method of claim 12 , wherein estimating the 3D depth of the vehicle comprises:

determining 3D points corresponding to the 2D crossing points; and

estimating the 3D depth based on the 3D points and the position of the vehicle.

14. An apparatus, comprising:

a memory; and

a processor communicatively coupled to the memory,

wherein the memory and/or the processor are configured to:

determine a 2D bounding box (2DBB) associated with a vehicle within a camera frame; and

estimate a 3DBB associated with the vehicle based on the 2DBB and lane information, the 3DBB comprising a first box and a second box, the first box associated with a first portion of the vehicle and the second box associated with a second portion of the vehicle,

wherein the lane information comprises information on road geometry of one or more lanes within a field of view of one or more sensors of an ego vehicle,

wherein the first and second boxes together bound an entirety of the vehicle,

wherein the 2DBB is a truncated BB associated with a partial view of the vehicle, and

wherein to estimate the 3DBB, the memory and the processor are configured to:

select a pseudo vehicle based on the 2DBB;

generate one or more pseudo 3DBBs associated with the pseudo vehicle on corresponding one or more lanes; and

select one of the pseudo 3DBBs based a comparison of the one or more pseudo 3DBBs and the 2DBB.

15. The apparatus of claim 14 , wherein the 2DBB bounds the vehicle.

16. The apparatus of claim 14 , wherein the first and second boxes share a common edge.

17. The method of claim 15 , wherein to select the pseudo vehicle, the memory and the processor are configured to:

determine a class of the vehicle based on the 2DBB; and

select the pseudo vehicle from a plurality of class vehicles based on the determined class of the vehicle.

18. The method of claim 15 , wherein to select one of the pseudo 3DBBs, the memory and the processor are configured to:

for each pseudo 3DBB, project the pseudo 3DBB such that a vertical side of the projected 3DBB is aligned with a vertical side of the 2DBB;

for each pseudo 3DBB, compare the projected 3DBB with the 2DBB; and

select one of the pseudo 3DBBs based on the comparison.

19. The method of claim 14 , wherein to generate the one or more pseudo 3DBBs, the memory and the processor are configured to:

determine the one or more lanes within the camera frame; and

generate the one or more 3DBB of the pseudo vehicle placed in the one or more lanes.

20. The apparatus of claim 19 , wherein the memory and the processor are configured to determine the one or more lanes from processing prior camera frames and/or from a lane map.

21. The apparatus of claim 19 , wherein the memory and the processor are configured to determine the one or more lanes from a lane map.

22. An apparatus, comprising:

a memory; and

a processor communicatively coupled to the memory,

wherein the memory and/or the processor are configured to:

determine a 2D bounding box (2DBB) associated with a vehicle within a camera frame; and

estimate a 3DBB associated with the vehicle based on the 2DBB and lane information, the 3DBB comprising a first box and a second box, the first box associated with a first portion of the vehicle and the second box associated with a second portion of the vehicle,

wherein the lane information comprises information on road geometry of one or more lanes within a field of view of one or more sensors of an ego vehicle,

wherein the first and second boxes together bound an entirety of the vehicle,

wherein the memory and the processor are further configured to update the estimated 3DBB by:

projecting the estimated 3DBB onto the 2DBB;

extracting the first portion of the vehicle from the 2DBB based on the projected estimated 3DBB;

determining lane boundaries of a lane the vehicle is located within the camera frame;

determining a position of the vehicle within the lane of the vehicle;

estimating a 3D depth of the vehicle; and

updating the estimated 3DBB based on the 3D depth.

23. The apparatus of claim 22 , wherein determining the lane boundaries comprises lane curve fitting on 2D lane vertices.

24. The apparatus of claim 23 , wherein the 2D lane vertices are obtained from lane information, the lane information included in a lane map or derived from processing prior camera frames.

25. The apparatus of claim 22 , wherein determining the position of the vehicle within the lane comprises:

determining 2D crossing points between the lane boundaries and a line extended from a bottom side of the first portion of the vehicle; and

determining the position of the vehicle as a position of a center of the vehicle relative to a midpoint of the 2D crossing points.

26. The apparatus of claim 25 , wherein estimating the 3D depth of the vehicle comprises:

determining 3D points corresponding to the 2D crossing points; and

estimating the 3D depth based on the 3D points and the position of the vehicle.