A data fusion strategy for distributed multi-source heterogeneous sensors
TL;DR
This study systematically analyzes the failure mechanisms of multimodal sensors and proposes an innovative distributed data fusion strategy that improves target recognition accuracy, meets real-time perception latency constraints, and significantly enhances the robustness of environmental perception systems.
Abstract
Multi-source heterogeneous sensor data fusion serves as the core technology for environmental perception systems in intelligent vehicles, playing a decisive role in ensuring driving safety. To address the interference issues of sensor detection features in complex environments, this study systematically analyzes the failure mechanisms of multimodal sensors and proposes an innovative distributed data fusion strategy. The method establishes a collaborative framework of wavelet analysis and federated filtering for data preprocessing, and develops an environment-adaptive feature-level fusion algorithm with real-time calibration drift compensation. By dynamically evaluating the effectiveness of multi-sensor features, it enables intelligent interference identification and fusion weight optimization in complex scenarios. Real-vehicle experiments demonstrate that under extreme environmental conditions such as rain, fog, and strong light, the proposed solution improves target recognition accuracy, meets real-time perception latency constraints, and significantly enhances the robustness of environmental perception systems.