NVIDIA GPU: can process images from 12 cameras

 

    NVIDIA Corporation of the United States held a variety of demonstrations at the company's GPU technology developer conference "GPU Technology Conference (GTC) 2015" using the "DRIVE PX" autonomous driving development platform (evaluation board) for ADAS (advanced driver assistance systems) and autonomous driving.

    DRIVE PX is equipped with two "Tegra X1" processor ICs, which feature high processing performance for automotive use. Tegra X1 is equipped with a 256-core GPU based on the "Maxwell" architecture.

    At the exhibition, NVIDIA demonstrated how to connect 12 monocular cameras equipped on the vehicle body to DRIVE PX to display images. The commentator said, "General surround view systems can process images from 4 cameras, but with our GPU performance, it is more than enough to process images from 12 cameras. It can also stitch images obtained by each camera together and identify objects in the image." In addition, the demonstration vehicle is equipped with 12 cameras, including cameras for shooting farther ahead, cameras for shooting closer ahead, cameras inside the car to monitor driver fatigue, and cameras that replace side mirrors.

    In addition, NVIDIA also demonstrated an automatic parking system using four monocular cameras and DRIVE PX, emphasizing that the images of the four monocular cameras can be seamlessly stitched together for the surround view system.

    In addition, in order to recognize the images of four monocular cameras in three dimensions (3D), "SfM (Structure from Motion)" technology is used. That is, the camera position is moved, the angle of view is changed, multiple images are acquired, and then the three-dimensional model of the photographed object is restored by comparing the images. In this way, the situation around the vehicle can be grasped, thereby realizing automatic parking. NVIDIA simulated and demonstrated this technology at the venue.