A deep learning-based radar and vision integrated machine lane line automatic acquisition method

Abstract

This invention relates to the field of intelligent transportation technology, and in particular to a method for automatic lane line acquisition using a radar-visual integrated machine based on deep learning. The method includes: acquiring video images and radar point clouds; establishing a radar-visual mapping; determining pixel coordinates and pixel confidence levels; constructing visual lane lines; generating radar center lines; dividing effective and ineffective sections; calculating lane width; obtaining corrected lane lines; generating virtual lane lines; and obtaining multi-lane lines based on the corrected lane lines and virtual lane lines. This invention effectively solves the problems in existing technologies, such as lane line detection breaks due to environmental influences in visual perception, radar center line fitting deviations due to insufficient accuracy of traditional trajectory clustering, visual failures that cannot be compensated for due to the lack of lane line-level complementarity in target-level fusion, and systematic drift of fused coordinates due to the inability of offline calibration to adapt to equipment deformation.

Images