There is growing demand for high-definition maps to improve the stability of current autonomous driving technology. However, the current process for building high-definition maps involves a high proportion of manual labor for digitizing and structural editing, making it difficult to maintain road conditions that frequently change. Moreover, as the quality of a high-definition map varies with the skill of the person creating it, it is difficult to achieve consistency. Accordingly, in this study, we propose a methodology that extracts areas of road lane markings from point clouds acquired by mobile mapping systems. The methodology uses a deep learning model to predict the color type of road lane markings, then automatically generates a high-definition map layer. Positioning accuracy and vector structuring tests were performed to verify the usability of the road lane marking vector data generated using the proposed methodology. In the positioning accuracy test, the maximum error for the horizontal and vertical positions was within 0.2 m and the root mean square error at the 95% confidence level was within 0.1 m for the original and generated vector data. In the vector structuring test, both study areas showed a high structuring accuracy of 85% or more.