Enhancing oriented object detection in RADAR through temporal affine transformation analysis for autonomous driving

Abstract

Recent advancements in autonomous driving have seen a growing interest in object detection research using RADAR. RADAR offers the advantages of cost-efficiency and high reliability in adverse weather conditions compared to cameras or LiDAR. However, the low angular resolution of RADAR increases uncertainty in object detection. To overcome this, previous studies have sought to enhance performance through deep learning based temporal modeling and sensor fusion. While effective for low-speed vehicle RADAR, these methods have limitations at high speeds due to inadequate consideration of the data acquisition speed of the sensor. To address this issue, we propose a method that explicitly improves temporal modeling through the analysis of affine transformations between consecutive frames. This approach, applied to feature fusion, distinctly models moving vehicles and stationary backgrounds, significantly enhancing object detection performance. The efficacy of our model has been demonstrated on the RADIATE that is RADAR dataset.