Estimating road lane geometry using lane marker observations

摘要

Aspects of the disclosure relate generally to detecting the edges of lane lines. Specifically, a vehicle driving on a roadway may use a laser to collect data for the roadway. A computer may process the data received from the laser in order to extract the points which potentially reside on two lane lines defining a lane. The extracted points are used by the computer to determine a model of a left lane edge and a right lane edge for the lane. The model may be used to estimate a centerline between the two lane lines. All or some of the model and centerline estimates, may be used to maneuver a vehicle in real time and also to update or generate map information used to maneuver vehicles.

主权项

1. A computer implemented method comprising:
accessing sensor data collected for a roadway, the sensor data including data points having location information for objects; determining an initial estimate for a location of an edge of a lane marker bounding a lane; generating a set of possible estimates for the location of the edge based on an offset distance and the initial estimate; for each estimate of the set of possible estimates, determining a likelihood value indicative of how likely the estimate is to be the location of the edge; selecting a most likely estimate of the set of possible estimates based on the likelihood values; fitting, by a processor, the sensor data points to a model describing a location and shape of the edge; and determining a representation of the edge based on the model.