Prevents overshoot and undershoot at edge portions by excluding pixels with significant brightness changes from the reference pool during smoothing.
A model generation device adds situation information to object recognition data for machine learning.
Selective image acquisition reduces data transmission volume by filtering intersections lacking prominent landmarks.