Trajectory Prediction of Neighboring Vehicles via Periodic Beaconing with Inaccurate GPS Data

Predicting trajectories of neighboring vehicles accurately is essential for safe driving in Advanced Driving Assistance Systems (ADAS). Conventional approaches for predicting trajectory rely upon an expensive image sensor (i.e., lidar). However, high cost but limited sensing range of lidar makes installing a trajectory prediction system on existing vehicles impractical. Using two functionalities provided by a smartphone, i.e., periodic beaconing and GPS positioning could address issues of conventional approaches. However, modules for wireless communications and GPS are limited in achieving accurate prediction in their current form: transmission error and inaccurate GPS data. In this paper, we propose a novel scheme for predicting neighbors' trajectories using periodic beaconing and GPS, and show that our scheme is feasible to trajectory prediction. To address the limitations of the wireless and GPS modules, our scheme integrates Long Short-Term Memory (LSTM) with two strategies: 1) Tailoring Multi-band Transmission to Prediction (TMTP) and 2) Environment-Aware Positioning (EAP). To improve prediction accuracy even with the imperfect input, our scheme employs interactive LSTM model. To our knowledge, this is the first to enable predicting the neighbors' trajectories using periodic beaconing with inaccurate GPS data. Experimental results using real traces demonstrate that our scheme is accurate in predicting neighbors' trajectories.