GTC Europe 2016 - Multiple Deep Neural Network Demo

Опубликовано 21 октября 2016, 12:25

Using NVIDIA DriveWorks, we demonstrate the use of three different Deep Neural Network trained by our research team using a single front facing camera and running sequentially in real-time. The first network detects drivable areas in front of the car and, at the same time, labels the limits of the area as curb/wall (white), vehicle (yellow) or pedestrian (red). The second network is a car detector which outputs 3D bounding boxes that describe the size and 3D orientation of the car. The last network is a lane detector that is able to mark egolane and neighboring lanes.