Heterogeneous multi-threaded visual odometry in autonomous driving
A system and method for performing visual odometry is disclosed. In aspects, the system implements methods to generate an image pyramid based on an input image received. A refined pose prior information representing a location and orientation of the autonomous vehicle can be generated based on one or more images of the image pyramid. One or more seed points can be selected from the one or more images of the image pyramid. One or more refined seed representing the one or more seed points with added depth values can be generated. One or more scene points can be generated based on the one or more refined seed points. A point cloud can be generated based on the one or more scene points.
1. A computer implemented method for performing visual-odometry, the method comprising:
identifying, by an on-board computing device of an autonomous vehicle (AV), a failure of the AV's connection with a guidance system;
based on the identification of the failure of the AV's connection with the guidance system:
generating, by a graphics processing unit (GPU), an image pyramid based on an input image received, wherein the input image received represents an image of an environment in which an autonomous vehicle is being operated;
generating, by the GPU, a refined pose prior information representing a location and orientation of the autonomous vehicle based on one or more images of the image pyramid;
selecting, by the GPU, one or more seed points from the one or more images of the image pyramid, the one or more seed points representing pixel locations within one or more images of the image pyramid representing estimations of where an object is likely located;
generating, by the GPU, one or more refined seed points representing the one or more seed points with added depth values;
generating, by a central processing unit (CPU), one or more scene points based on the one or more refined seed points;
generating, by the CPU, a point cloud based on the one or more scene points; and
navigating, by the on-board computing device and using the point cloud, the AV to an area of safety for it to be shut down.
2. The method of claim 1 , further comprising generating, by the GPU or the CPU, one or more first scene points and first keyframes as initial values for performing the visual-odometry, based on performing an initialization process utilizing the image pyramid.
3. The method of claim 2 , wherein the initialization process comprises:
selecting one or more first seed points from one or more first images of a first image pyramid;
generating one or more first refined seed points representing the one or more first seed points with added depth values;
generating one or more first scene points based on the one or more refined first seed points; and
generating the first keyframes based on the one or more first scene points.
4. The method of claim 1 , further comprising generating, by the CPU, a keyframe based on the input image received.
5. The method of claim 1 , further comprising decoding, by the GPU, the input image received to generate a decoded image.
6. The method of claim 5 , wherein generating the image pyramid comprises resizing the decoded image to generate a set of images, wherein the set of images comprise the images of the image pyramid.
7. The method of claim 1 , wherein selecting the one or more seed points from the one or more images of the image pyramid further comprises:
dividing the one or more images of the image pyramid into block regions; and
selecting a pixel in each block region of the block regions with the largest gradient as a seed point for each block region.
8. A non-transitory computer readable medium including instructions for causing one or more processors to perform operations for performing visual-odometry, the operations comprising:
identifying a failure of an autonomous vehicle's(AV) connection with a guidance system;
based on the identification of the failure of the AV's connection with the guidance system:
generating, by a graphics processing unit (GPU), an image pyramid based on an input image received, wherein the input image received represents an image of an environment in which an autonomous vehicle is being operated;
generating, by the GPU, a refined pose prior information representing a location and orientation of the autonomous vehicle based on one or more images of the image pyramid;
selecting, by the GPU, one or more seed points from the one or more images of the image pyramid, the one or more seed points representing pixel locations within one or more images of the image pyramid representing estimations of where an object is likely located;
generating, by the GPU, one or more refined seed points representing the one or more seed points with added depth values;
generating, by a central processing unit (CPU), one or more scene points based on the one or more refined seed points;
generating, by the CPU, a point cloud based on the one or more first scene points; and
navigating, using the point cloud, the AV to an area of safety for it to be shut down.
9. The non-transitory computer readable medium of claim 8 , wherein the operations further comprise generating, by the GPU or the CPU, one or more first scene points and first keyframes as initial values for performing the visual-odometry, based on performing an initialization process utilizing the image pyramid.
10. The non-transitory computer readable medium of claim 9 , wherein the initialization process comprises:
selecting one or more first seed points from one or more first images of a first image pyramid;
generating one or more first refined seed points representing the one or more first seed points with added depth values;
generating one or more first scene points based on the one or more refined first seed points; and
generating the first keyframes based on the one or more first scene points.
11. The non-transitory computer readable medium of claim 8 , the operations further comprising generating, by the CPU, a keyframe based on the input image received.
12. The non-transitory computer readable medium of claim 8 , the operations further comprising decoding, by the GPU, the input image received to generate a decoded image.
13. The non-transitory computer readable medium of claim 12 , wherein generating the image pyramid comprises resizing the decoded image to generate a set of images, wherein the set of images comprise the images of the image pyramid.
14. The non-transitory computer readable medium of claim 8 , wherein selecting the one or more seed points from the one or more images of the image pyramid further comprises:
dividing the one or more images of the image pyramid into block regions; and
selecting a pixel in each block region of the block regions with the largest gradient as a seed point for each block region.
15. A computing system of an autonomous vehicle (AV) for performing visual-odometry comprising:
a storage unit to store instructions;
an on-board computing device coupled to the storage unit configured to:
identifying a failure of the AV's connection with a guidance system; and
based on identifying the failure of the AV's connection with the guidance system:
generate, with a graphics processing unit (GPU), an image pyramid based on an input image received, wherein the input image received represents an image of an environment in which an autonomous vehicle is being operated,
generate, with the GPU, a refined pose prior information representing a location and orientation of the autonomous vehicle based on one or more images of the image pyramid,
select, with the GPU, one or more seed points from the one or more images of the image pyramid, the one or more seed points representing pixel locations within one or more images of the image pyramid representing estimations of where an object is likely located,
generate, with the GPU, one or more refined seed points representing the one or more seed points with added depth values, and
generate, with a central processing unit (CPU), one or more scene points based on the one or more refined seed points;
generate, with the CPU, a point cloud based on the one or more scene points; and
navigate, using the point cloud, the AV to an area of safety for it to be shut down.
16. The computing system of claim 15 , wherein the GPU or the CPU is further configured to generate one or more first scene points and first keyframes as initial values for performing the visual-odometry, based on performing an initialization process utilizing the image pyramid.
17. The computing system of claim 16 , wherein the GPU or the CPU is further configured to perform the initialization process by:
selecting one or more first seed points from one or more first images of a first image pyramid;
generating one or more first refined seed points representing the one or more first seed points with added depth values;
generating one or more first scene points based on the one or more refined first seed points; and
generating the first keyframes based on the one or more first scene points.
18. The computing system of claim 15 , wherein the CPU is further configured to generate a keyframe based on the input image received.
19. The computing system of claim 15 , wherein the GPU is further configured to decode the input image received to generate a decoded image.
20. The computing system of claim 19 , wherein the GPU is further configured to generate the image pyramid by resizing the decoded image to generate a set of images, wherein the set of images comprise the images of the image pyramid.