Enhancing basic roadway-intersection models using high intensity image data

摘要

Systems and methods are provided that may optimize basic models of an intersection in a roadway with high intensity image data of the intersection of the roadway. More specifically, parameters that define the basic model of the intersection in the roadway may be adjusted to more accurately define the intersection. For example, by comparing a shape of the intersection predicted by the basic model with extracted curbs and lane boundaries from elevation and intensity maps, the intersection parameters can be optimized to match real intersection-features in the environment. Once the optimal intersection parameters have been found, roadgraph features describing the intersection may be extracted.

主权项

1. A method comprising:
receiving, using a computing device, parameterized intersection data indicative of an intersection in a roadway of an environment, wherein the parameterized intersection data includes a plurality of parameters that define the intersection in the environment; based on the plurality of parameters, determining a first configuration of the intersection, wherein the first configuration of the intersection includes one or more of predicted curb locations, predicted lane locations, or predicted pedestrian-control line locations; receiving map data based on a detection of the roadway in the environment by sensors on a vehicle that traverses the environment, wherein the map data includes one or more of candidate curb locations of the intersection, candidate lane locations of the intersection, or candidate pedestrian-control lines of the intersection; performing a first optimization of the predicted curb locations to the candidate curb locations to minimize correspondence error between the predicted curb locations and the candidate curb locations for optimal curb locations; performing a second optimization of the predicted lane locations to the candidate lane locations to minimize correspondence error between the predicted lane locations and the candidate lane locations for optimal lane locations; performing a third optimization of the predicted pedestrian-control lines to the candidate pedestrian-control lines to minimize correspondence error between the predicted pedestrian-control line locations and the candidate pedestrian-control line locations for optimal pedestrian-control line locations; based on one or more of the first optimization, the second optimization, and the third optimization, determining a second configuration of the intersection that includes the optimal curb locations, the optimal lane locations, and the optimal pedestrian-control line locations; and based on the second configuration of the intersection, navigating an autonomous vehicle through the intersection.