Pavement lane marking detection using matched filter

Abstract This paper proposes novel algorithms to detect pavement lane markings on two-dimensional (2D) laser images via three stages. First, a new matched filter is developed to generate strongest responses at pavement lane markings. Second, a novel hybrid thresholding method is proposed to preserve strong responses while eliminating weak responses. Last, the Shape Examination is conducted to verify if the geometric shapes of detected objects are similar to the pattern of pavement lane markings, resulting in a higher confidence level. The experimental results on six real pavement sections demonstrated that the proposed algorithms achieved high F-measures greater than 92% for pavement lane markings in excellent or fair conditions. In addition, the F-measures were higher than 82% even for the two sections with lane markings in severe conditions, which should reflect the efficiency of the proposed algorithms.