Accurate lane marker geometry is important for autonomous vehicles to navigate roads safely and efficiently. Map patches provide valuable data for estimating lane marker geometry but need to be aligned before they can be used to estimate the lane marker geometry. Traditional feature-based approaches for aligning map patches are unsuitable for map patches that include only lane markers, which typically lack distinctive features. This disclosure describes techniques for aligning map patches, specifically aligning lane markers on road geometries. Per the technique, correspondence between lane markers is determined, focusing on smooth curves without distinct features. Map patch alignment technique combines sensor observations from different vehicles to infer the lane marker geometry and update a digital map. An objective function is defined that includes a regularization term on the magnitude of rotation and translation. The techniques involve estimating the rotation angle directly using a nonlinear solver, forming correspondences between lane markers, and using the area between lane markers as a distance metric. The techniques bring together multiple map patches that each represent a segment of the roadway and position the patches correctly relative to each other.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.