Visual localization against a prior map
A system and method for performing visual localization is disclosed. In aspects, the system implements methods to generate a global point cloud, the global point cloud representing a plurality of point clouds. The global point cloud can be mapped to a prior map information to locate a position of an autonomous vehicle, the prior map information representing pre-built geographic maps. The position of the autonomous vehicle can be estimated based on applying sensor information obtained from sensors and software of the autonomous vehicle to the mapped global point cloud.
1. A computer implemented method for performing visual localization, the method comprising:
generating, by one or more computing devices, a global point cloud, the global point cloud representing a plurality of point clouds;
mapping, by the one or more computing devices, the global point cloud to a prior map information to locate a position and orientation of an autonomous vehicle, the prior map information representing pre-built geographic maps;
estimating, by the one or more computing devices, the position and orientation of the autonomous vehicle based on applying sensor information obtained from sensors and software of the autonomous vehicle to the mapped global point cloud;
receiving, by the one or more computing devices, an initial estimate of the position and orientation of the autonomous vehicle; and
mapping, by the one or more computing devices, the global point cloud to the prior map information based on the initial estimate to provide an enhanced determination of a surrounding of the autonomous vehicle.
2. The method of claim 1 , further comprising generating the global point cloud based on:
projecting one or more scene points of a point cloud of the plurality of point clouds onto one or more keyframes associated with a further point cloud; and
performing an optimization process to minimize a photometric error to obtain the global point cloud, the photometric error being for the one or more scene points in the one or more keyframes.
3. The method of claim 2 , further comprising checking, by the one or more computing devices, for duplicate one or more scene points across the one or more keyframes by checking whether projections of the one or more scene points are close in distance to further scene points from a keyframe image of the one or more keyframes.
4. The method of claim 2 , further comprising performing the optimization process to generate a set of optimized scene point depths and keyframes with cross camera matches.
5. The method of claim 1 , wherein the mapping of the global point cloud to the prior map information is performed by utilizing an Iterative Closest Point (ICP) algorithm to perform the mapping.
6. The method of claim 1 , wherein the initial estimate of the position and orientation of the autonomous vehicle is obtained based on sensor data collected and processed as the autonomous vehicle is operated.
7. A non-transitory computer readable medium including instructions for causing one or more processors to perform operations for performing visual localization, the operations comprising:
generating a global point cloud, the global point cloud representing a plurality of point clouds;
mapping the global point cloud to a prior map information to locate a position and orientation of an autonomous vehicle, the prior map information representing pre-built geographic maps;
estimating the position and orientation of the autonomous vehicle based on applying sensor information obtained from sensors and software of the autonomous vehicle to the mapped global point cloud;
receiving an initial estimate of the position and orientation of the autonomous vehicle; and
mapping the global point cloud to the prior map information based on the initial estimate to provide an enhanced determination of a surrounding of the autonomous vehicle.
8. The non-transitory computer readable medium of claim 7 , wherein the operations further comprise generating the global point cloud based on:
projecting one or more scene points of a point cloud of the plurality of point clouds onto one or more keyframes associated with a further point cloud; and
performing an optimization process to minimize a photometric error to obtain the global point cloud, the photometric error being for the one or more scene points in the one or more keyframes.
9. The non-transitory computer readable medium of claim 8 , wherein the operations further comprise checking for duplicate one or more scene points across the one or more keyframes by checking whether projections of the one or more scene points are close in distance to further scene points from a keyframe image of the one or more keyframes.
10. The non-transitory computer readable medium of claim 8 , wherein the operations further comprise performing the optimization process to generate a set of optimized scene point depths and keyframes with cross camera matches.
11. The non-transitory computer readable medium of claim 7 , wherein the mapping of the global point cloud to the prior map information is performed by utilizing an Iterative Closest Point (ICP) algorithm to perform the mapping.
12. The non-transitory computer readable medium of claim 7 , wherein the initial estimate of the position and orientation of the autonomous vehicle is obtained based on sensor data collected and processed as the autonomous vehicle is operated.
13. A computing system for performing visual localization comprising:
a storage unit to store instructions;
a control unit, coupled to the storage unit, configured to process the stored instructions to:
generate a global point cloud the global point cloud representing a plurality of point clouds;
map the global point cloud to a prior map information to locate a position and orientation of an autonomous vehicle, the prior map information representing pre-built geographic maps;
estimate the position and orientation of the autonomous vehicle based on applying sensor information obtained from sensors and software of the autonomous vehicle to the mapped global point cloud;
receive an initial estimate of the position and orientation of the autonomous vehicle; and
map the global point cloud to the prior map information based on the initial estimate to provide an enhanced determination of a surrounding of the autonomous vehicle.
14. The computing system of claim 13 , wherein the control unit is further configured to generate the global point cloud by:
project one or more scene points of a point cloud of the plurality of point clouds onto one or more keyframes associated with a further point cloud; and
perform an optimization process to minimize a photometric error to obtain the global point cloud, the photometric error being for the one or more scene points in the one or more keyframes.
15. The computing system to claim 14 , wherein the control unit is further configured to check for duplicate one or more scene points across the one or more keyframes by checking whether projections of the one or more scene points are close in distance to further scene points from a keyframe image of the one or more keyframes.
16. The computing system to claim 14 , wherein the control unit is further configured to perform the optimization process to generate a set of optimized scene point depths and keyframes with cross camera matches.
17. The computing system of claim 13 , wherein the initial estimate of the position and orientation of the autonomous vehicle is obtained based on sensor data collected and processed as the autonomous vehicle is operated.