A robust extraction method for railway using sequential line detection

Visual extraction of railway is widely used in automatic driving and obstacle detection. At present, there are many research results on lane line detection. However, there is a big difference between lane line detection and railway extraction. Rail extraction faces many challenges, such as complex environments and complex rail shapes. This paper introduces a railway extraction method based on sequential line detection. In this method, picture is divided into multiple stages from bottom to top. According to proposed matching score, select lines that are most matching to railway in each stage are selected. Then, all detected lines are linked as final extracted railway, which makes proposed method is effective to straight railway and curved railway. Experiments have proved that proposed method remains accurate and stable in different environments.