Extraction of single-pixel-width single edge using dynamic thresholding for lane detection on structured roads

Abstract

Traditional edge extraction algorithms usually suffer from drawbacks such as difficult selection of thresholds, serious interferences outside the lane, and redundant pixels at the lane edges when used for the lane detection on structured roads. This study proposed a new edge extraction algorithm by using dynamic thresholding. More precisely, for each row of an edge-enhanced image, by scanning from middle towards both left and right, two sets of dynamic thresholds were built by collecting three dominant pixel values along each direction, and a unique edge position was marked simultaneously at each side by using the corresponding threshold set. This thresholding strategy can provide the most concise yet adequate edge points for forming a single-pixel-width single edge of a lane marking, with particular improvements in the presence of interferences from illuminations and shadows, and with significant reduction of the computation load for the subsequent lane detection, as is demonstrated through experiments.