A method for detecting the three lane markers in a set of continuous two-lane highway images using linear approximation is presented. The system is designed for accurate and robust extraction of this data as well as high processing speed.
After edge detection, vertical vibration in the image is measured, and the edges of the lane markers are approximated using line fitting. The vanishing point and line gradients are estimated, assuming that all lines cross at the vanishing point. A least squares method is used with conditions derived from the time and space relationships between the measured lines, the vertical vibration, and the results from the previous frame.
High accuracy was proven by quantitative evaluation using simulated images. Both robustness and the practicality of real time video rate processing were then confirmed through experiment using VTR real road images.
抄録全体を表示