Route to Time and Time to Route: Travel Time Estimation from Sparse Trajectories

Due to the rapid development of Internet Things (IoT) technologies, many online web apps (e.g., Google Map and Uber) estimate travel time trajectory data collected by mobile devices. However, in reality, complex factors, such as network communication energy constraints, make multiple trajectories at a low sampling rate. In this case, paper aims resolve problem estimation (TTE) route recovery sparse scenarios, which often leads uncertain label between continuously sampled GPS points. We formulate an inexact supervision training has coarsely grained labels jointly solve tasks TTE recovery. And we argue that both two are complementary each other model-learning procedure hold relation: more precise can lead better inference for routes (Time $$\rightarrow $$ Route), turn, resulting accurate (Route Time). Based on assumption, propose EM algorithm alternatively inferred through weak E step retrieve based estimated M trajectories. conducted experiments three real-world datasets demonstrated effectiveness proposed method.