CoFormerNet: A Transformer-Based Fusion Approach for Enhanced Vehicle-Infrastructure Cooperative Perception

Area of Science:

• Computer Science
• Robotics
• Artificial Intelligence

Background:

• Autonomous driving systems rely heavily on accurate environmental perception.
• Cooperative perception, integrating vehicle and infrastructure data, offers enhanced situational awareness.
• Existing methods struggle with communication delays and spatial misalignment in cooperative perception.

Purpose of the Study:

• To introduce CoFormerNet, a novel framework for robust vehicle-infrastructure cooperative perception.
• To address challenges like communication delays and spatial misalignment in fusing sensor data.
• To improve the performance of 3D object detection in autonomous driving.

Main Methods:

• CoFormerNet utilizes a consistent structure for both vehicle and infrastructure perception branches.
• It integrates a temporal aggregation module for handling time-varying data.
• Spatial-modulated cross-attention fuses intermediate features at two distinct stages.

Main Results:

• CoFormerNet demonstrated superior performance compared to existing methods on the DAIR-V2X and V2XSet datasets.
• The framework achieved state-of-the-art results in 3D object detection.
• The proposed fusion strategy effectively mitigates issues of communication delay and spatial misalignment.

Conclusions:

• CoFormerNet offers a significant advancement in vehicle-infrastructure cooperative perception.
• The framework's design effectively handles real-world challenges in data fusion.
• CoFormerNet paves the way for more reliable and safer autonomous driving systems.