Hybrid Sensor Fusion Approach for Robust Perception

Modern automated vehicles are equipped with a variety of sensors, including cameras, radar, and LiDAR, each with unique strengths and weaknesses. To leverage these complementary capabilities, perception algorithms fuse inputs from multiple sensors to build a comprehensive environmental representation. Among fusion-based approaches, deep fusion algorithms, which combine features extracted from raw sensor signals, demonstrate superior performance on various dataset benchmarks. However, these approaches are inherently vulnerable to sensor failures, as their detection pipelines rely on inputs from multiple sensor modalities. To address this limitation, we propose a novel architecture that combines a single-sensor detection pipeline with a deep fusion-based object detector. Our approach improves robustness in scenarios with sensor signal loss while maintaining high performance under normal operating conditions.