A New Semantic Segmentation Technique for Interference Mitigation in Automotive Radar

Recent autonomous vehicles and Advanced Driver Assistance Systems (ADAS) are equipped with several sensing technologies, including cameras, LiDAR, radar, and ultrasonic. Due to its exceptional features, radar is increasingly utilized in a range of ADAS applications. Unfortunately, this increases the likelihood of radar-to-radar interference, which hinders radar functionality. Numerous research studies have investigated interference mitigation using various traditional signal processing or deep learning techniques. This paper presents a new technique utilizing the U-Net deep neural network (DNN) model for interference mitigation via semantic segmentation in such ADAS scenarios. By comparing the performance of the proposed model to previously published deep-learning-based approaches, our new model has demonstrated promising improvements based on standard evaluation criteria.