A novel approach for accelerated and accurate spatial conflation of connected vehicle data on GPUs and its application to real-time statewide traffic state estimation

There is a growing desire among transportation agencies to consider augmenting traditional traffic data collection with high-resolution data streaming directly from vehicles connected to the internet. Merging this dataset with traditional data streams has the potential to improve decision making for transportation systems maintenance, operations and safety. Critical use cases such as shockwave estimation, incident detection, crash risk prediction require near real-time spatial conflation to roads, and other infrastructure mounted traffic sensors. Here, we show a novel conflation algorithm, FastConflate, that combines high-resolution spatial grid cells and novel bottom-up and ring-based conflation technique for fast, accurate spatial matching of large datasets in real-time. Compared to existing algorithms, FastConflate exhibits unparalleled speed and precision, spatial indexing of 50 million points in a mere 9 ms–making it 14,000×, 17,000×, and 106,000× faster than the H3, S2, and geohash algorithms respectively. Moreover, on a standard desktop GPU, it conflates 50 million points to road networks in just 4.2 s, making it faster than cloud versions of Heavy AI-GPU database, PostgreSQL, and Google Big Query by 400×, 344×, and 41× respectively. In addition, we provide a pipeline and show that FastConflate can be applied to real-time transportation intelligence by applying statewide traffic state estimates for speed and volume. Given that FastConflate overcomes the challenges of existing algorithms by being highly accurate at intersections and interchanges, the proposed approach offers hope of implementing real-time transportation intelligence decisions which will lead to enhanced system safety and efficacy to state-wide transportation networks.