Ego-Vehicle Speed Correction for Automotive Radar Systems Using Convolutional Neural Networks

Area of Science:

• Automotive Engineering
• Computer Vision
• Signal Processing

Background:

• Growing demand for advanced driver-assistance systems (ADAS) necessitates robust automotive radar.
• Current radar systems face challenges with angular performance and mounting variations impacting detection range.

Purpose of the Study:

• To propose a novel convolutional neural network model, AVSD Net, for accurate ego-vehicle speed estimation using automotive radar data.
• To detail preprocessing and postprocessing techniques for precise vehicle speed correction.

Main Methods:

• Development of the AVSD Net model utilizing convolutional neural networks.
• Inputting range-velocity spectrum data into the AVSD Net model.
• Implementing detailed preprocessing and postprocessing for speed correction.

Main Results:

• The AVSD Net model effectively estimates ego-vehicle speed, independent of radar angular performance and mounting angle.
• The method reduces detection range loss without specialized beam requirements.
• Preprocessing and postprocessing achieve accurate speed correction with reduced model complexity.

Conclusions:

• The proposed AVSD Net model offers an efficient solution for ego-vehicle speed estimation in automotive radar systems.
• The method's robustness and reduced complexity make it suitable for embedded systems.
• This speed correction technique significantly enhances safety features in ADAS applications.