Object tracking by generating velocity grids
A computer-implemented method is provided for creating a velocity grid using a simulated environment for use by an autonomous vehicle. The method may include simulating a road scenario with simulated objects. The method may also include recording image data collected from a camera sensor, the image data comprising a first 2D image frame comprising the simulated objects made up of a plurality of pixels. The method may also include identifying a first 3D point on the first simulated object in a 3D view of the simulated road scenario, wherein the first 3D point corresponds to the first pixel in the first 2D image frame. The method may also include generating a velocity of the first point based upon a velocity of the first simulated object, and projecting the velocity back into the first 2D image frame. The method may further include encoding the velocity for the first pixel prior to a simulated movement of the simulated object with respect to a second pixel after the simulated movement of the simulated object in a 2D velocity grid.
1. A computer-implemented method for creating a velocity grid using a simulated environment for use by an autonomous vehicle, the method comprising:
simulating a road scenario with simulated objects;
recording image data collected from a camera sensor, the image data comprising a first two-dimensional (2D) image frame comprising the simulated objects made up of a plurality of pixels at a first location, and a second 2D image frame comprising the simulated objects made up of a plurality of pixels at a second location;
identifying a first three-dimensional (3D) point on a first simulated object in a 3D view of the simulated road scenario, wherein the first 3D point corresponds to the first pixel in the first 2D image frame;
identifying a second 3D point on the first simulated object in a 3D view of the simulated road scenario, wherein the second 3D point corresponds to a first pixel in the second 2D image frame;
determining a distance between the first pixel in the first 2D image frame at the first location and the first pixel in the second 2D image frame at the second location during a simulated movement of the simulated object;
generating a velocity of the first point based upon the determined distance;
projecting the velocity back into the first 2D image frame; and
encoding the velocity for the first pixel prior to a simulated movement of the simulated object with respect to a second pixel after the simulated movement of the simulated object in a 2D velocity grid.
2. The computer-implemented method of claim 1 , the identifying a first three-dimensional (3D) point on the first simulated object comprising:
recording simulated LiDAR rays reflecting off the first simulated object; and
determining the first 3D point that corresponds to the first pixel in the first two-dimensional (2D) image frame based upon intersection of the LiDAR rays and the first simulated object in the simulated road scenario.
3. The computer-implemented method of claim 1 , the generating a velocity of the first three-dimensional (3D) point comprising:
simulating a movement of the first simulated object in the simulated road scenario in a period of time;
calculating the velocity of the first 3D point based upon the velocity of the first simulated object; and
recording a second two-dimensional (2D) image frame after the simulated movement of the first simulated object, wherein a second 3D point corresponds to a second pixel in the second 2D image frame.
4. The computer-implemented method of claim 3 , further comprising projecting the second three-dimensional (3D) point into the second two-dimensional (2D) image frame, wherein the second pixel corresponds to the same portion of the first simulated object as the first pixel.
5. The computer-implemented method of claim 1 , wherein the velocity of the first simulated object is known in the simulated road scenario.
6. The computer-implemented method of claim 1 , wherein a period of time for the simulated movement of the first simulated object is 100 ms between the first two-dimensional (2D) image frame and the second 2D image frame.
7. The computer-implemented method of claim 1 , further comprising training the autonomous vehicle using a machine learning algorithm including the two-dimensional (2D) velocity grid of the simulated objects in the simulated environment.
8. A system for creating a velocity grid using a simulated environment for use by an autonomous vehicle, the system comprising:
one or more processors;
a computer-readable medium comprising instructions stored therein, which when executed by one or more processors, cause the processors to:
simulate a road scenario with simulated objects, record image data collected from a camera sensor, the image data comprising a first two-dimensional (2D) image frame comprising the simulated objects made up of a plurality of pixels at a first location, and a second 2D image frame comprising the simulated objects made up of a plurality of pixels at a second location, identify a first three-dimensional (3D) point on the first simulated object in a 3D view of the simulated road scenario, wherein the first 3D point corresponds to the first pixel in the first 2D image frame, identify a second 3D point on the first simulated object in a 3D view of the simulated road scenario, wherein the second 3D point corresponds to a first pixel in the second 2D image frame, determine a distance between the first pixel in the first 2D image frame at the first location and the first pixel in the second 2D image frame at the second location during a simulated movement of the simulated object, generate a velocity of the first point based upon the determined distance, project the velocity back into the first 2D image frame, and
encode the velocity for the first pixel prior to a simulated movement of the simulated object with respect to a second pixel after the simulated movement of the simulated object in a 2D velocity grid.
9. The simulation system of claim 8 , wherein the instructions to identify a first three-dimensional (3D) point on the first simulated object further cause the processors to:
record simulated LiDAR rays reflecting off the first simulated object; and
determine the first 3D point that corresponds to the first pixel in the first two-dimensional (2D) image frame based upon intersection of the LiDAR rays and the first simulated object in the simulated road scenario.
10. The simulation system of claim 8 , wherein the instructions to generate a velocity of the first three-dimensional (3D) point further cause the processors to:
simulate a movement of the first simulated object in the simulated road scenario in a period of time;
calculate the velocity of the first 3D point based upon the velocity of the first simulated object; and
record a second two-dimensional (2D) image frame after the simulated movement of the first simulated object, wherein a second 3D point corresponds to a second pixel in the second 2D image frame.
11. The simulation system of claim 10 , wherein the instructions further cause the processors to project the second three-dimensional (3D) point into the second two-dimensional (2D) image frame, wherein the second pixel corresponds to the same portion of the first simulated object as the first pixel.
12. The simulation system of claim 8 , wherein the velocity of the first simulated object is known in the simulated road scenario.
13. The simulation system of claim 8 , wherein a period of time for the simulated movement of the first simulated object is 100 ms between the first two-dimensional (2D) image frame and the second 2D image frame.
14. The simulation system of claim 8 , wherein the instructions further cause the processors to train the autonomous vehicle using a machine learning algorithm including the two-dimensional (2D) velocity grid of the simulated objects in the simulated environment.
15. A non-transitory computer-readable storage medium comprising:
a storage configured to store instructions;
a processor configured to execute the instructions and cause the processor to:
simulate a road scenario with simulated objects, record image data collected from a camera sensor, the image data comprising a first two-dimensional (2D) image frame comprising the simulated objects made up of a plurality of pixels at a first location, and a second 2D image frame comprising the simulated objects made up of a plurality of pixels at a second location, identify a first three-dimensional (3D) point on the first simulated object in a 3D view of the simulated road scenario, wherein the first 3D point corresponds to the first pixel in the first 2D image frame, identify a second 3D point on the first simulated object in a 3D view of the simulated road scenario, wherein the second 3D point corresponds to a first pixel in the second 2D image frame, determine a distance between the first pixel in the first 2D image frame at the first location and the first pixel in the second 2D image frame at the second location during a simulated movement of the simulated object, generate a velocity of the first point based upon the determined distance, project the velocity back into the first 2D image frame, and
encode the velocity for the first pixel prior to a simulated movement of the simulated object with respect to a second pixel after the simulated movement of the simulated object in a 2D velocity grid.
16. The non-transitory computer-readable storage medium of claim 15 , wherein the instructions to identify a first three-dimensional (3D) point on the first simulated object further cause the processors to:
record simulated LiDAR rays reflecting off the first simulated object; and
determine the first 3D point that corresponds to the first pixel in the first two-dimensional (2D) image frame based upon intersection of the LiDAR rays and the first simulated object in the simulated road scenario.
17. The non-transitory computer-readable storage medium of claim 15 , wherein the instructions to generate a velocity of the first three-dimensional (3D) point further causes the processors to:
simulate a movement of the first simulated object in the simulated road scenario in a period of time;
calculate the velocity of the first 3D point based upon the velocity of the first simulated object; and
record a second two-dimensional (2D) image frame after the simulated movement of the first simulated object, wherein a second 3D point corresponds to a second pixel in the second 2D image frame.
18. The non-transitory computer-readable storage medium of claim 17 , wherein the instructions further cause the processors to project the second three-dimensional (3D) point into the second two-dimensional (2D) image frame, wherein the second pixel corresponds to the same portion of the first simulated object as the first pixel.
19. The non-transitory computer-readable storage medium of claim 15 , wherein the velocity of the first simulated object is known in the simulated road scenario.
20. The non-transitory computer-readable storage medium of claim 15 , wherein a period of time for the simulated movement of the first simulated object is 100 ms between the first two-dimensional (2D) image frame and the second 2D image frame.