The Landmark Detection module detects landmarks through a streamed video as input. It involves loading the NVIDIA proprietary landmark detection deep neural network: MapNet. MapNet runs inference using the loaded network, and post-processes the network output to receive an array of landmark polylines.
There are four types of landmarks currently supported: lane markings, poles, road markings, and intersection markings.
Additional landmark types will be supported in the future.
The Road mark classifier module classifies road markings on camera images using output from landmark detector. NVIDIA proprietary DNN named RoadMarkNet will return the type of detected road marking.