About 360 camera setups for self driving cars
Google is investigating ways to better project spherical video and is using cubes in different ways.
If we consider each side of the polygon as a potential target for a camera, what kind of polygons can we devise that have a lot more resolution where we care about (front and back at 4K) and not a lot where we don’t care much (WVGA for the sky)?
And there are several ways to create polygons that would let us then put a lot of tightly packed voxels so we can segment entire scenes:
My favorite one is this one:
Look how beautiful it is when it is applied to the earth:
But we are not limited to waterman polyhedra, we can use ANY polygon according to Erik Demaine from MIT:
The researchers’ algorithm designs crease patterns for producing any polyhedron — that is, a 3-D surface made up of many flat facets. Computer graphics software, for instance, models 3-D objects as polyhedra consisting of many tiny triangles. “Any curved shape you could approximate with lots of little flat sides,” Demaine explains.
This means we can map any shape we create to a 4K 16:9 screen and then feed it to neural networks, it does not have to be a cube.
More than a hundred years ago
While working on the telephone, Bell mentioned to Watson that their next project would be a flying machine. On his honeymoon, he told his wife Mabel that he dreamed of flying machines with telephones attached.
Like the Wrights and other aviation pioneers, Bellchose to test light, wind-supported kite and glider designs before attempting risky human-powered flight trials.
I understand Google sticking to cubes for general purpose video, but for self driving cars, should we be dreaming a little bit more?