Abstract: High- quality datasets have become the cornerstone of intelligentization in the transportation industry, particularly in the context of smart highways, which integrate multi- source data such as ETC gantries, video surveillance, radar detection, geomagnetic coils, and meteorological monitoring. However, the temporal references, sampling intervals, spatial representations, and field semantics of these heterogeneous data sources are often inconsistent, leading to misalignment when directly applied to fusion analysis or model training. To address this challenge, this paper proposes a method for constructing high- quality smart highway datasets based on spatiotemporal consistency, which is refined into three dimensions: temporal consistency, spatial consistency, and semantic consistency. The construction framework encompassed standardized access, spatiotemporal alignment, consistency verification, and iterative evaluation and optimization, with a specific focus on aligning spatiotemporal dependencies, building inter- data relationships, and conducting consistency verification and assessment across multi- source data. The proposed approach aimed to resolve the core issues of data interpretation and trustworthy fusion, thereby providing a feasible methodology for building high- quality datasets for smart highway systems.