Mounting Angle Prediction for Automotive Radar Using Complex-Valued Convolutional Neural Network

Area of Science:

• Engineering
• Computer Science
• Artificial Intelligence

Background:

• Automotive radar mounting angle misalignment critically degrades object detection and tracking accuracy in ADAS.
• This degradation compromises the safety and overall performance of autonomous driving systems.

Purpose of the Study:

• To introduce a novel model, AutoRAD-Net, for accurate azimuth alignment detection in automotive radars.
• To address the challenges posed by mounting angle offsets in radar systems.

Main Methods:

• Utilized a complex-valued convolutional neural network (CV-CNN) architecture.
• Trained and validated the model with mounting angle offsets ranging from -3° to +3°.
• Evaluated the model's generalization capability on unseen offsets.

Main Results:

• AutoRAD-Net achieved azimuth alignment errors of no more than 0.15° across tested offsets.
• Demonstrated reliable prediction accuracy for previously unseen angle offsets, such as -1.7°.
• The model effectively learns physical properties from complex-valued radar data.

Conclusions:

• AutoRAD-Net offers a practical and effective solution for automotive radar azimuth alignment.
• The predicted offsets can facilitate physical alignment or be integrated into compensation algorithms.
• This advancement enhances radar reliability and data interpretation accuracy in ADAS applications.