Real-Time End-to-End Prediction of Driving Routes

16  December  2021, 12:30 PM at Room 206 And via zoom

Abstract:

For a long time now, location data has not been the sole preserve of navigation devices, and nowadays such user data is collected from almost all electronic devices. The wealth of data and the ever increasing ability of processing such vast quantities, offers unprecedented information to analyze human mobility, cultivating an extensive variety of applications in location-based services such as intelligent transportation systems and smart cities. In recent years, many studies have been conducted into predicting the next location the user is expected to reach, based on a history of previous locations. Existing route-prediction methods have difficulty coping with a large number of trips, all the more so if they are spread across a wide geographical area.

The research presented in this paper introduces a new framework that allows real-time end-to-end prediction of a user route. First, preliminary processing is carried out during which all the data is divided into routes. Second, a linear and accurate approximation of the dynamic time warping algorithm, which is 50 to 150 times faster than standard DTW, is applied to divide the various trips into identical routes. A long short-term memory (LSTM) neural network model is applied to capture the geographic and the temporal dependencies, and is used to predict the route of a real-time trip. We tested our method on a real-life GPS traces dataset with high temporal and spatial resolution, and the experimental results demonstrate that our method has a better performance than the compared models.
 

Bio:

Moran Sorka is an M.Sc student in the Department of Industrial Engineering at Tel Aviv University, under the supervision of Prof. Irad Ben-Gal.
She holds a B.Sc in Industrial Engineering and Information Systems from The Technion.
She works as a Researcher Data Scientist in Cognyte (formerly Verint company).