A fast and Accurate Sketch Method for Estimating User Similarities over Trajectory Data

TL;DR

This paper introduces a novel bidirectional RNN-based weighted trajectory reconstruction method incorporating neural arithmetic logic units, significantly improving accuracy and robustness in urban vehicle trajectory estimation despite GNSS signal issues.

Contribution

The paper presents a new trajectory reconstruction algorithm that combines multi-source data fusion with NALU-enhanced deep learning for improved urban vehicle trajectory accuracy.

Findings

Reduces RMSE in trajectory reconstruction

Demonstrates robustness in complex urban environments

Outperforms existing methods in accuracy and reliability

Abstract

In a complex urban environment, due to the unavoidable interruption of GNSS positioning signals and the accumulation of errors during vehicle driving, the collected vehicle trajectory data is likely to be inaccurate and incomplete. A weighted trajectory reconstruction algorithm based on a bidirectional RNN deep network is proposed. GNSS/OBD trajectory acquisition equipment is used to collect vehicle trajectory information, and multi-source data fusion is used to realize bidirectional weighted trajectory reconstruction. At the same time, the neural arithmetic logic unit (NALU) is introduced into the trajectory reconstruction model to strengthen the extrapolation ability of the deep network and ensure the accuracy of trajectory prediction, which can improve the robustness of the algorithm in trajectory reconstruction when dealing with complex urban road sections. The actual urban road section was selected for testing experiments, and a comparative analysis was carried out with existing methods. Through root mean square error (RMSE, root-mean-square error) and using Google Earth to visualize the reconstructed trajectory, the experimental results demonstrate the effectiveness and reliability of the proposed algorithm.