Visual-inertial positional awareness for autonomous and non-autonomous mapping
The described positional awareness techniques employing visual-inertial sensory data gathering and analysis hardware with reference to specific example implementations implement improvements in the use of sensors, techniques and hardware design that can enable specific embodiments to provide positional awareness to machines with improved speed and accuracy.
1. A system including:
one or more processors coupled to memory loaded with computer instructions to provide guidance information to guide a mobile platform, which instructions, when executed on the processors, implement actions comprising:
selecting from a plurality of sets of image data, one or more sets of image data to include in a point cloud of features, wherein image data includes feature points determined from captured image data and pose information that includes a location and an orientation of a camera that captured the image data;
for selected sets of image data, (a) triangulating new feature points using feature points of sets of image data previously added to the point cloud of features to obtain feature points in a coordinate system of the mobile platform, and (b) aligning coordinates of the feature points in the point cloud of features to a coordinate system having a z-axis aligned with gravity;
creating a multilayered hybrid point grid from feature points selected for the point cloud of features, including at least one layer of a multilayered 2D occupancy grid; and
providing the at least one layer of a multilayered 2D occupancy grid to a process that plans a path of travel to a location.
2. The system of claim 1 , further implementing actions comprising refining the point cloud of features by:
selecting sets of at least some of image data and pose information of the point cloud of features;
reprojecting 3D points from the point cloud of features onto a 2D image plane using the pose information;
calculating image coordinates for the 3D points as reprojected on the 2D image plane and comparing the image coordinates as calculated for the 3D points as reprojected on the 2D image plane to actual coordinates of corresponding observed points obtained directly from the image data selected to compute a plurality of re-projection errors; and
minimizing a weighted sum of the re-projection errors as computed.
3. The system of claim 2 , wherein the selecting further includes selecting image data and pose information in which IMU readings for static images is [0,0, 1 g] thereby aligning z-axis of the image coordinates with gravity.
4. The system of claim 2 , wherein the minimizing further includes constraining a pose to moving along a plane when the mobile platform is mounted on a vehicle with planar motion.
5. The system of claim 4 , wherein the constraining includes applying a pose moving along a plane constraint to a cost function.
6. The system of claim 2 , further including refining the multilayered hybrid point grid by: using ray tracing, determining cells in each layer of the multilayered 2D occupancy grid to update based upon one or more points in the point cloud of features as refined by:
determining a ray between a point in the point cloud of features as refined and a corresponding point located by a cell of an x-y layer of the multilayered 2D occupancy grid such that the ray is substantially parallel to a direction of gravity in a coordinate system of the multilayered 2D occupancy grid; and
marking the cell as containing an object;
finding cells along the ray between the point in the point cloud of features as refined and the corresponding point in the x-y layer of the multilayered 2D occupancy grid and marking the cells found as empty; and
finding cells between the point in the point cloud of features as refined and a point from which the image data is captured and marking the cells as empty.
7. The system of claim 1 , further including providing pose information with respect to the point cloud of features to track position of the mobile platform that captured images used to build the point cloud of features.
8. The system of claim 1 , wherein the triangulating further includes one selected from (i) triangulating new feature points across images from a current set of image data, (ii) triangulating new feature points across images from two different sets of image data, wherein two different sets of image data are not necessarily in sequence, (iii) triangulating new feature points from images in sets of image data chosen based upon a criterion.
9. The system of claim 1 , wherein creating a multilayered hybrid point grid further includes:
populating at least one layer of a 2D occupancy grid that corresponds to an x-y layer including at least some of the feature points in the point cloud of features, with points from the point cloud of features within a height range using ray tracing from an observed location of a point aligned to a corresponding point in the multilayered 2D occupancy grid and a location of a corresponding point reprojected on the layer of the multilayered 2D occupancy grid; and
finding cells along a ray between the observed location of the point as aligned and the corresponding point reprojected on the layer and marking the any cells found as empty.
10. A method of providing guidance information to guide a mobile device including:
selecting from a plurality of sets of image data, one or more sets of image data to include in a point cloud of features, wherein image data includes feature points determined from captured image data and pose information that includes a location and an orientation of a camera that captured the image data;
for selected sets of image data, (a) triangulating new feature points using feature points of sets of image data previously added to the point cloud of features to obtain feature points in a coordinate system of the mobile device, and (b) aligning coordinates of the feature points in the point cloud of features to a coordinate system having a z-axis aligned with gravity;
creating a multilayered hybrid point grid from feature points selected for the point cloud of features, including at least one layer of a multilayered 2D occupancy grid; and
providing the at least one layer of a multilayered 2D occupancy grid to a process that plans a path of travel to a location.
11. The method of claim 10 , further including refining the point cloud of features by:
selecting sets of at least some of image data and pose information of the point cloud of features;
reprojecting 3D points from the point cloud of features onto a 2D image plane using the pose information;
calculating image coordinates for the 3D points as reprojected on the 2D image plane and comparing the image coordinates as calculated for the 3D points as reprojected on the 2D image plane to actual coordinates of corresponding observed points obtained directly from the image data selected to compute a plurality of re-projection errors; and
minimizing a weighted sum of the re-projection errors computed.
12. The method of claim 11 , wherein the selecting further includes selecting image data and pose information in which IMU readings for static images is [0,0,1 g] thereby aligning z-axis of the image coordinates with gravity.
13. The method of claim 11 , wherein the minimizing further includes constraining a pose to moving along a plane when the mobile device is mounted on a vehicle with planar motion.
14. The method of claim 13 , wherein the constraining includes applying a pose moving along a plane constraint to a cost function.
15. The method of claim 11 , further including refining the multilayered hybrid point grid by: using ray tracing, determining cells in each layer of the multilayered 2D occupancy grid to update based upon one or more points in the point cloud of features as refined by:
determining a ray between a point in the point cloud of features as refined and a corresponding point located by a cell of an x-y layer of the multilayered 2D occupancy grid such that the ray is substantially parallel to a direction of gravity in a coordinate system of the multilayered 2D occupancy grid; and
marking the cell as containing an object;
finding cells along the ray between the point in the point cloud of features as refined and the corresponding point in the x-y layer of the multilayered 2D occupancy grid and marking the cells found as empty; and
finding cells between the point in the point cloud of features as refined and a point from which the image data is captured and marking the cells as empty.
16. The method of claim 10 , further including providing pose information with respect to the point cloud of features to track position of the mobile device that captured images used to build the point cloud of features.
17. The method of claim 10 , wherein the triangulating further includes one selected from (i) triangulating new feature points across images from a current set of image data, (ii) triangulating new feature points across images from two different sets of image data, wherein two different sets of image data are not necessarily in sequence, (iii) triangulating new feature points from images in sets of image data chosen based upon a criterion.
18. The method of claim 10 , wherein creating a multilayered hybrid point grid further includes:
populating at least one layer of a 2D occupancy grid that corresponds to an x-y layer including at least some of the feature points in the point cloud of features, with points from the point cloud of features within a height range using ray tracing from an observed location of a point aligned to a corresponding point in the multilayered 2D occupancy grid and a location of a corresponding point reprojected on the layer of the multilayered 2D occupancy grid; and
finding cells along a ray between the observed location of the point as aligned and the corresponding point reprojected on the layer and marking the any cells found as empty.
19. A non-transitory computer readable storage medium impressed with computer program instructions to provide guidance information to guide a mobile device, which instructions, when executed on a processor, implement a method comprising:
selecting from a plurality of sets of image data, one or more sets of image data to include in a point cloud of features, wherein image data includes feature points determined from captured image data and pose information that includes a location and an orientation of a camera that captured the image data;
for selected sets of image data, (a) triangulating new feature points using feature points of sets of image data previously added to the point cloud of features to obtain feature points in a coordinate system of the mobile device, and (b) aligning coordinates of the feature points in the point cloud of features to a coordinate system having a z-axis aligned with gravity;
creating a multilayered hybrid point grid from feature points selected for the point cloud of features, including at least one layer of a multilayered 2D occupancy grid; and
providing the at least one layer of a multilayered 2D occupancy grid to a process that plans a path of travel to a location.
20. The non-transitory computer readable storage medium of claim 19 , wherein creating a multilayered hybrid point grid further includes:
populating at least one layer of a 2D occupancy grid that corresponds to an x-y layer including at least some of the feature points in the point cloud of features, with points from the point cloud of features within a height range using ray tracing from an observed location of a point aligned to a corresponding point in the multilayered 2D occupancy grid and a location of a corresponding point reprojected on the layer of the multilayered 2D occupancy grid; and
finding cells along a ray between the observed location of the point as aligned and the corresponding point reprojected on the layer and marking the any cells found as empty.