- Note
- SW Release Applicability: This module is available in NVIDIA DRIVE Software releases.
About This Module
The Localization module determines the precise 6-DOF position and orientation of a vehicle within an HD map. NVIDIA uses landmark localization to refine and improve the estimated vehicle pose from Egomotion; landmark localization involves comparing real-time landmark detections (such as lane lines, road boundaries, road signs, or poles) with their counterparts in an HD map to determine an exact pose. Highly accurate placement is necessary for tasks such as path planning and maneuvering using HD map data.
The Localization module takes three types of measurements as input. First, the module requires an approximate, absolute vehicle position in WGS84 and orientation w.r.t. ENU at that position. Second, the module needs a relative transformation from the vehicle at the previous time of calling the module to the current time. Third and finally, the module needs detected landmarks. Localization is intended to improve the output of the Egomotion module: the absolute and relative position and orientation data should come from egomotion. Landmark detections can be computed with the Landmark Perception and the Object Perception modules. The localization module maintains an internal history of localized poses and detection matching, and requires that each call be at a monotonically increasing time. Localization results are more accurate after a "burn in" phase of about 50 calls to the module.
The output of the Localization module consists of three distinct parts. First, the module outputs an absolute position in WGS84 and orientation w.r.t. ENU at the position. Next, the output specifies the relative pose to the closest road segment in the HD map. This section consists of a 3 element ID specifying the closest road segment, the relative pose w.r.t. an ENU coordinate frame about the origin of the closest road segment (refer to Maps for more information on road segments), and a flag indicating whether the relative pose is valid. Lastly, the output provides a measure of uncertainty for the position and orientation.
- Note
- Localization performs undistortion of the input lane detections based on an FTheta camera model that is provided at initialization time.
The following figure depicts HD map components projected into the camera field of view after localization (left), as well as the vehicle rig position rendered inside the HD map after localization (right).
