Trajectories know where map is wrong: an iterative framework for map-trajectory co-optimisation

The low map quality has been a persistent problem which is usually caused by the belated map update. Although the recent research on map inference/update enables timely map update through the use of trajectory data, the update quality is still far from being practically useful due to the trajectory inaccuracy. In this work, we propose an iterative map-trajectory co-optimisation framework which refines the traditional map inference/update results by considering their contribution to the quality improvement on both map and trajectory map-matching results. In each iteration, we propose two respective scores to measure the credibility and influence of each road update and refine the map and map-matching result accordingly. Meanwhile, we quantify the quality of map and trajectory-matching results so that the goal of our iterative co-optimisation is to maximise the overall quality result. Additionally, to accelerate the iterative process, we introduce an R-tree-based spatial index to avoid unnecessary map-matching. Overall, our framework supports most of the existing map inference/update methods and significantly improves the quality of their update result with affordable overhead. We conduct extensive experiments on real-world datasets of different scales. The results show the significant quality improvement over the state-of-the-art map update methods while the efficiency stays competitive.