Data Fusion with Split Covariance Intersection for Cooperative Perception

Cooperative Perception is an emergent technology that profits from the exchanged perception information between vehicles. However, the inconsistency resulting from the reuse of the same information is a main issue that arises. In this paper, we focus on the study of the Split Covariance Intersection Filter (SCIF), a method capable of handling both independent and arbitrarily correlated estimates and observation errors. We are interested in its use in a Cooperative Perception application to incorporate information coming from other vehicles, which may or may not have been tracked, in a generic tracking solution. A simple case study is first presented to build a deep understanding of the filter tuning, then real experiments carried out with three vehicles equipped with GNSS, camera, LiDAR and High Definition (HD) map features are reported to study how a full tracking architecture relying on SCIF behaves in a real-world situation.