Comparison of Curve Representations for Memory-Efficient and High-Precision Map Generation

High-precision maps provide essential and detailed information for automated vehicles, especially about the individual lanes of a road. Here, the memory efficiency of curve representations is a critical aspect to limit the amount of data to store and process. There are many specific approaches in literature that generate spline-based maps from sensor data, however only a few evaluate the memory requirement. Furthermore, different algorithms are developed for each specific spline type. In this contribution, a generic optimization-based framework for the generation of spline curves with arbitrary degree and continuity is proposed by adapting an algorithm from the field of computer aided design. Here, the continuity of the spline is explicitly optimized which enhances the approximation capabilities. The method is evaluated for two datasets with a total length of 34.22 km. Comparative approaches are outperformed in terms of memory efficiency and robustness while the proposed method yields an average memory requirement of less than 3 byte/m.