Appreciation and Application of Your Work in Late Fusion for Collaborative Perception

[maryemfadili]

First of all, thank you for your incredible contributions to collaborative perception. Your repository has been a true goldmine for us over the past two years.

Thanks to the insights and tools provided here, we developed two late fusion algorithms—an area we found to be underexplored in the current literature despite its practical importance. In real-world deployments, onboard and offboard perception systems are often developed by different manufacturers. This makes:

Early fusion approaches challenging due to high bandwidth requirements, and

Deep fusion approaches impractical, as they often assume full knowledge of the internal architectures of both systems—something rarely available in real-world conditions.

Motivated by these challenges, we focused on late fusion, designing sensor- and detector-agnostic algorithms that run in real time (<10 ms per frame).

Our work led to three accepted papers:

🔹 Weighted Least Squares Multi-Detection Fusion and Kalman Filter Based Tracking for Collaborative Perception Systems, accepted at ICNSC 2025
📄 PDF

🔹 A Late Collaborative Perception Framework for 3D Multi-Object and Multi-Source Association and Fusion, accepted at ICRAS 2025
📄 PDF

🔹 Uncertainty-Aware Late Fusion Framework for Collaborative Perception, accepted at IRCE 2025
📄 PDF

We hope this highlights how impactful your open-source work has been in supporting downstream research and real-world applications. Thank you again!