Abstract: Currently, the extraction and updating of road network has been one of the key factors affecting urban construction and development. At the same time, with the continuous development of driverless technology, the construction of high-precision road network is one of the key research contents of many scholars. Existing road network extraction methods based on track data have little semantic information for grid feature mining. Therefore, this paper proposes a road network extraction method of track data coupled with multi-level grid features. Firstly, the original track data is preprocessed, and multi-level grid features of the track data are calculated based on grid, including track similarity, grid track point density, etc. Then,based on multi-level grid features,the random forest model is used for feature training,and the key grid is classified and recognized. Finally,the key grid is extracted based on morphology. In this paper,the walking track data is used for model training,and the vehicle track data is used for model migration verification.The experimental results show that the proposed method has better performance than other road network extraction methods.

Key words: road extraction, trajectory data, grid, morphology, random forest