Calibration-Free Roadside BEV Perception with V2X-Enabled Vehicle Position Assistance

Area of Science:

• Computer Vision
• Autonomous Driving Systems
• Robotics

Background:

• Autonomous driving systems require comprehensive environmental awareness, often achieved through vehicle-based bird's eye view (BEV) perception.
• Existing methods rely on precise camera calibration or depth estimation, which can introduce inaccuracies.
• Roadside perception offers a complementary approach to enhance situational awareness.

Purpose of the Study:

• To develop a calibration-free roadside BEV perception architecture for autonomous driving.
• To improve the robustness of BEV perception against real-world communication and positioning uncertainties.
• To evaluate the proposed architecture against existing methods using a relevant dataset.

Main Methods:

• Utilized elevated roadside cameras and vehicle positions transmitted via cellular vehicle-to-everything (C-V2X) communication.
• Developed a calibration-free approach, independent of intrinsic camera parameters.
• Simulated practical issues like C-V2X delay, packet loss, and positioning noise by injecting random coordinate noise and varying data proportions.

Main Results:

• The proposed architecture demonstrated superior performance in roadside BEV perception compared to calibration-based and calibration-free baselines.
• The system effectively handles uncertainties inherent in C-V2X communication and vehicle positioning.
• Experimental validation on the DAIR-V2X dataset confirmed the effectiveness of the proposed method.

Conclusions:

• A novel, calibration-free roadside BEV perception architecture enhances autonomous driving capabilities.
• The C-V2X integrated system offers a robust solution for environmental awareness, mitigating common real-world data issues.
• This approach provides a promising direction for improving the safety and reliability of autonomous vehicles.