Evaluation of an Uncertainty-Aware Late Fusion Algorithm for Multi-Source Bird's Eye View Detections Under Controlled Noise

2025-07-04

AI Summary

Evaluating multi-sensor bird’s-eye view (BEV) detection fusion performance independently of detector errors remains challenging. To address this, we propose a noise-controllable, systematic evaluation framework and design UniKF—a Kalman filter-based, uncertainty-aware unified post-fusion algorithm. Our framework decouples detector errors from fusion performance via controlled noise injection into BEV feature maps or detections. UniKF explicitly models sensor-specific uncertainties (e.g., calibration and measurement noise) and temporal synchronization offsets, enabling robust cross-sensor fusion under realistic conditions. Experiments across varying noise levels demonstrate that UniKF reduces localization and orientation errors by approximately 3× and size estimation error by 2×, while maintaining near-perfect detection accuracy (99.5%–100% precision and recall). These results significantly outperform state-of-the-art fusion methods, validating both the efficacy of our evaluation framework and the robustness of UniKF under uncertainty.

Technology Category

• Intelligent Robots: State Estimation
• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Reliable multi-source fusion is crucial for robust perception in autonomous systems. However, evaluating fusion performance independently of detection errors remains challenging. This work introduces a systematic evaluation framework that injects controlled noise into ground-truth bounding boxes to isolate the fusion process. We then propose Unified Kalman Fusion (UniKF), a late-fusion algorithm based on Kalman filtering to merge Bird's Eye View (BEV) detections while handling synchronization issues. Experiments show that UniKF outperforms baseline methods across various noise levels, achieving up to 3x lower object's positioning and orientation errors and 2x lower dimension estimation errors, while maintaining nearperfect precision and recall between 99.5% and 100%.

Problem

• Evaluating multi-source fusion performance without detection errors
• Improving Bird's Eye View detection accuracy with noise
• Handling synchronization issues in late-fusion algorithms

Innovation

• Unified Kalman Fusion for BEV detections
• Controlled noise injection in evaluation
• Handles synchronization issues effectively