if you will notice they use only one camera, and the vector map is low resolution, and they are breaking even the low resolution into a 1d line to make decisions. and they only use the vectors when there is a whole other layer of easily recognizable data. that layer is artifical, yet easily obtainable. it is new/discarded pixels. if you watch something moving the motion vectors can tell you roughly how big it is, and which direction it is moving, but new/discarded pixels can give a presice outline of foreground objects.
also with higher resolution in the vector maps you could aproach a true 3d description of scene... especially with the addition of a second camera.
the vision system will use an established resolution standard for safe operation of a vehicle. i will make sure i have enough resolution that the system could pass a standard DMV eye test. i'll find out the values, and convert to angular resolution, then meet or excede that with the cameras.
After doing some research I have found that 20/40 is the minimum eyesight required to legaly drive (in california though it might vary state to state... this is just preliminary anyway). Eyesight of 20/40 coresponds to an angular resolution of about 2 seconds of arc... 30 pixels per degree... 640x480 = 21x16 degrees... now to figure out how much coverage i need...
on a side note i think i know how i could quickly and easily turn vector flow information from 2 cameras and turn it into a polygon like 3d model.
first you examine each picture for vector flow continuity, by breaking the flow into large chunks by picking a spot, and working outwards until you hit a change in the way the flow vectors are changing. this coresponds to an edge. edges have endpoints, and connect enough endpoints and you get places where polygons could be placed. use triangulation of endpoints to locate in 3d space.
all the cameras I use will have to be calibrated meticulously. spacial coordinates for each pixel should be pre determined. perhaps a screen with a moving dot on it could be used to coordinate camera systems for spacial awareness.
