Spatial-Temporal Weighted Attention Model for Cooperative Vehicular Positioning System

Abstract

Multisensory cooperative localization has emerged as a promising approach to enhance positioning accuracy in vehicular ad hoc networks (VANETs). This article proposes a sensor fusion localization algorithm that integrates global navigation satellite system (GNSS), radar, and received signal strength indicator (RSSI) measurements to refine current localization using both present and historical measurements. To emphasize the differing levels of importance between historical and current measurements in cooperative localization, the proposed algorithm combines the capabilities of long short-term memory (LSTM) models for capturing temporal patterns, ensemble localization for enhancing neighboring estimations, and weighted attention mechanisms for effectively integrating information from both temporal and spatial domains. Extensive simulation results consistently demonstrate the superior localization performance of the proposed algorithm compared to state-of-the-art sensor fusion benchmark algorithms, including the derived Cramer-Rao lower bound (CRLB), when addressing a progressively increasing difficulty across two driving scenarios. The proposed cooperative localization algorithm improves localization error by at least 29% compared to original GNSS measurements.