1990 年 14 巻 49 号 p. 7-12
As the first step in the development of a vision sensor for an autonomous driving system a real-time, 33ms per frame, highway road image processing system is proposed. The system is aimed to extract by linear approximation the three lane markers from a set of continuous monochromatic video pictures of the road. These pictures were taken from a passenger car driving on a two lane highway at a speed of 100km per hour. The processing algorithm first measures vertical vibrations in the picture caused by the pitching motion of the car. Additionally, the interpolation of the results of the previous frame into the current frame calculation makes the system robust against noise and frequent breaks in the marking lines.