About This Module

This module provides the APIs to initialize, query, and release the NVIDIA proprietary road mark classification deep neural network: RoadMarkNet.

The data structures include:

dwRoadMarkNetModel: allows user to choose among RoadMarkNet models.
dwRoadMarkNetParams: in addition to dwRoadMarkNetModel, defines the RoadMarkNet model variant with specific precision and processor optimization to be loaded.

DriveWorks comes with the following RoadMarkNet model:

US only model: DW_ROADMARKNET_MODEL_US_V0: Supported on GPU

Inputs

RoadMarkNet APIs consume FP16 planar frames with a resolution of 960x604 pixels from AR0231 cameras (revision >= 4) and dwLandmarkDetection from LandMarkDetector. Internally, it uses road mark bounding box information from dwLandmarkDetection to crop input images and resize them to an image patch of 64x64. This image patch is then fed to the DNN as an input.

Outputs

RoadMarkNet currently supports 33 road mark classes in US:

BackGround
ArrowMarkStraight
ArrowMarkLeft
ArrowMarkRight
ArrowMarkStraightLeft
ArrowMarkStraightRight
ArrowMarkLeftRight
ArrowMarkStraightLeftRight
ArrowMarkMerge
SpeedLimitMark5MPH
SpeedLimitMark10MPH
SpeedLimitMark15MPH
SpeedLimitMark20MPH
SpeedLimitMark25MPH
SpeedLimitMark30MPH
SpeedLimitMark35MPH
SpeedLimitMark40MPH
SpeedLimitMark45MPH
SpeedLimitMark50MPH
SpeedLimitMark55MPH
SpeedLimitMark60MPH
SpeedLimitMark65MPH
SpeedLimitMark70MPH
SpeedLimitMark75MPH
SpeedLimitMark80MPH
TextMarkBike
TextMarkAhead
TextMarkStop
TextMarkSchool
TextMarkHump
BikeMark
HOVMark
HumpMark

Additional Information

Warning: The RoadMarkNet model currently deployed is mostly trained with United States road markings. It might experience reduced accuracy in other regions, particularly for road marking patterns not present in the US.

Relevant Tutorials

RoadMarkNet Workflow

APIs

RoadMarkNet Interface