Visual odometry for operating a movable device
A computer that includes a processor and a memory, the memory including instructions executable by the processor to select first side stereo images of first and second pairs of stereo images acquired at first and second time steps, respectively, and mask first and second first side stereo images, determine point features in masked first and second first side stereo images and determine line features in masked first and second first side stereo images. Matching point features in the masked first and second first side stereo images can be determined using a first attention/graph neural network (attn/GNN) based on keypoints determined based on the line features. Matching line features in the masked first and second first side stereo images can be determined using a second attn/GNN based on keypoints determined based on the line features. Three-dimensional (3D) locations in a scene can be determined by determining stereo disparity based on the matched point features included in the first first side stereo image and point features determined in a first second side stereo image of first and second stereo pairs of images and a 3D stereo camera pose can be determined by determining a perspective-n-point and line algorithm on the 3D locations.
1. A system, comprising:
a computer that includes a processor and a memory, the memory including instructions executable by the processor to:
select first side first and second stereo images of first and second pairs of stereo images acquired at first and second time steps by a first side camera of a stereo camera, respectively;
mask first and second first side stereo images;
determine point features in the masked first and second first side stereo images;
determine line features in the masked first and second first side stereo images;
determine matching point features in the masked first and second first side stereo images using a first attention/graph neural network (attn/GNN) based on keypoints determined based on the line features;
determine matching line features in the masked first and second first side stereo images using a second attn/GNN based on keypoints determined based on the line features;
determine three-dimensional (3D) locations in a scene by determining stereo disparity based on the matched point features included in the first first side stereo image and point features determined in a first second side stereo image of first and second pairs of stereo images; and
determine a 3D stereo camera pose by determining a perspective-n-points-and-line algorithm on the 3D locations.
2. The system of claim 1 , the instructions including further instructions to mask the first and second first side stereo images by determining first and second segmented images for the first and second first side stereo images by an image segmentor and masking the first and second first side stereo images based on the first and second segmented images, respectively.
3. The system of claim 1 , wherein masking the first and second first side stereo images removes dynamic objects and sky.
4. The system of claim 1 , the instructions including further instructions to determine the point features in the masked first and second first side stereo images by self-supervised machine learning using the first attn/GNN followed by a differentiable optimal transport processor.
5. The system of claim 1 , the instructions including further instructions to determine the line features in the masked first and second first side stereo images by self-supervised machine learning using the second attn/GNN followed by a differentiable optimal transport processor.
6. The system of claim 1 , the instructions including further instructions to determine the point features based on a Superpoint algorithm.
7. The system of claim 1 , the instructions including further instructions to determine the line features based on a SOLD2 algorithm.
8. The system of claim 1 , the instructions including further instructions to determine the keypoints by sampling the line features determined in the masked first and second first side stereo images.
9. The system of claim 1 , the instructions including further instructions to determine point features in the first second side stereo image of first and second pairs of stereo images based on correlating pixel neighborhoods surrounding point features in the first left stereo image.
10. The system of claim 1 , the instructions including further instructions to determine stereo disparity by determining a distance between point features in the first second side stereo image and the masked first first side stereo image and based on one or more of a baseline between stereo cameras, a focal distance, and a pixel scale.
11. The system of claim 1 , the instructions including further instructions to combine the 3D stereo camera pose with one or more of global navigation satellite system (GNSS) data, or inertial measurement unit (IMU) to determine a 3D vehicle pose based on an extended Kalman filter.
12. The system of claim 11 , the instructions including further instructions to determine a path polynomial based on the 3D vehicle pose, and operate a vehicle based on the path polynomial.
13. A method, comprising:
selecting first side first and second stereo images of first and second pairs of stereo images acquired at first and second time steps, respectively;
masking first and second first side stereo images;
determining point features in masked first and second first side stereo images;
determining line features in masked first and second first side stereo images;
determining matching point features in the masked first and second first side stereo images using a first attention/graph neural network (attn/GNN) based on keypoints determined based on the line features;
determining matching line features in the masked first and second first side stereo images using a second attn/GNN based on keypoints determined based on the line features;
determining three-dimensional (3D) locations in a scene by determining stereo disparity based on the matched point features included in the first first side stereo image and point features determined in a first second side stereo image of first and second pairs of stereo images; and
determining a 3D stereo camera pose by determining a perspective-n-points-and-line algorithm on the 3D locations.
14. The method of claim 13 , further comprising masking the first and second first side stereo images by determining first and second segmented images for the first and second first side stereo images by an image segmentor and masking the first and second first side stereo images based on the first and second segmented images, respectively.
15. The method of claim 13 , wherein masking the first and second first side stereo images removes dynamic objects and sky.
16. The method of claim 13 , further comprising determining the point features in the masked first and second first side stereo images by self-supervised machine learning using the first attn/GNN followed by a differentiable optimal transport processor.
17. The method of claim 13 , further comprising determining the line features in the masked first and second first side stereo images by self-supervised machine learning using the second attn/GNN followed by a differentiable optimal transport processor.
18. The method of claim 13 , further comprising determining the point features based on a Superpoint algorithm.
19. The method of claim 13 , further comprising determining the line features based on a SOLD2 algorithm.
20. The method of claim 13 , further comprising determining the keypoints by sampling the line features determined in the masked first and second first side stereo images.