Synthesizing three-dimensional visualizations from perspectives of onboard sensors of autonomous vehicles
Aspects of the disclosure provide for generating a visualization of a three-dimensional (3D) world view from the perspective of a camera of a vehicle. For example, images of a scene captured by a camera of the vehicle and 3D content for the scene may be received. A virtual camera model for the camera of the vehicle may be identified. A set of matrices may be generated using the virtual camera model. The set of matrices may be applied to the 3D content to create a 3D world view. The visualization may be generated using the 3D world view as an overlay with the image, and the visualization provides a real-world image from the perspective of the camera of the vehicle with one or more graphical overlays of the 3D content.
1. A method of creating a three-dimensional (3D) world view from a perspective of a camera of a vehicle, the method comprising:
receiving, by one or more processors, images of a scene captured by the camera of the vehicle and 3D content for the scene;
retrieving, by the one or more processors, calibration data for the camera of the vehicle;
identifying, by the one or more processors, a virtual camera model for the camera of the vehicle based on the calibration data and parameters associated with the camera of the vehicle; and
creating, by the one or more processors, the 3D world view using the identified virtual camera model.
2. The method of claim 1 , wherein creating the 3D world view includes:
generating, by the one or more processors, a set of matrices using the identified virtual camera model; and
applying, by the one or more processors, the set of matrices to the 3D content to create the 3D world view.
3. The method of claim 2 , further comprising:
generating, by the one or more processors, a visualization using the 3D world view as an overlay with an image of the images, wherein the visualization provides a real-world image from the perspective of the camera of the vehicle with one or more graphical overlays of the 3D content.
4. The method of claim 1 , wherein the parameters include one or more of focal length, aperture size, or position and rotation of the camera relative to the vehicle when the images were captured.
5. The method of claim 3 , wherein the set of matrices includes a view matrix.
6. The method of claim 5 , wherein the view matrix transforms from world-space coordinates into camera-space coordinates.
7. The method of claim 6 , wherein applying the set of matrices includes generating coordinates for the 3D content in a reference frame of the camera of the vehicle.
8. The method of claim 6 , wherein the view matrix is defined as a combination of a swap matrix, a camera extrinsics matrix, and a vehicle pose matrix.
9. The method of claim 8 , wherein the swap matrix is configured to perform a coordinate swap to transform from vehicle coordinates to 3D rendering coordinates.
10. The method of claim 8 , wherein the camera extrinsics matrix provides a 3D position and 3D rotation of the camera of the vehicle when the image of the images was captured.
11. The method of claim 8 , wherein the vehicle pose matrix provides rotation and position of the vehicle when the image of the images was captured.
12. The method of claim 2 , wherein the set of matrices includes a projection matrix.
13. The method of claim 12 , wherein the projection matrix is configured to transform 3D coordinates in a reference frame of the camera of the vehicle into clip-space coordinates.
14. The method of claim 12 , wherein the projection matrix is defined as a combination of a transformation matrix and an orthographic projection matrix.
15. The method of claim 14 , wherein the transformation matrix is configured to generate projected coordinates using a linear transformation of 3D content and distorting points from the perspective of the camera of the vehicle.
16. The method of claim 14 , wherein the orthographic projection matrix is configured to convert the projected coordinates into clip space coordinates.
17. The method of claim 1 , wherein the 3D world view incorporates 3D content from space outside of the identified virtual camera model.
18. The method of claim 1 , wherein the vehicle includes a plurality of cameras, the method further comprising:
identifying, by the one or more processors, a virtual camera model for each camera of the plurality the cameras; and
generating, by the one or more processors, a set of matrices for each camera of the plurality of cameras of the vehicle using the virtual camera model for the camera.
19. The method of claim 18 , wherein the set of matrices includes a view matrix that transforms from world-space coordinates into camera-space coordinates, the set of matrices including a projection matrix that transforms 3D coordinates in a reference frame of the camera of the vehicle.
20. A system comprising:
a vehicle including a camera; and
one or more processors configured to:
receive images of a scene captured by the camera of the vehicle and three-dimensional (3D) content for the scene;
retrieve calibration data for the camera of the vehicle;
identify a virtual camera model for the camera of the vehicle based on the calibration data and parameters associated with the camera of the vehicle; and
create a 3D world view using the identified virtual camera model.