Exploring Domain Shift on Radar-Based 3D Object Detection Amidst Diverse Environmental Conditions

TL;DR

This study investigates domain shift in 4D radar-based 3D object detection under varying weather and road conditions. It conducts cross-domain experiments using two datasets (K-Radar and Bosch-Radar) and two detectors (RTNH and SECOND) to quantify performance gaps with fixed test splits; finds substantial domain shifts across weather (notably snow) and road types, with shifts being dataset-dependent and not reliably mitigated by simply increasing data size or changing architectures. It identifies radar point-cloud generation as a key source of dataset-specific domain gaps and emphasizes the need for diverse data collection across conditions. It proposes that domain adaptation techniques are essential to robust radar-based perception for autonomous driving in real-world variability.

Abstract

The rapid evolution of deep learning and its integration with autonomous driving systems have led to substantial advancements in 3D perception using multimodal sensors. Notably, radar sensors show greater robustness compared to cameras and lidar under adverse weather and varying illumination conditions. This study delves into the often-overlooked yet crucial issue of domain shift in 4D radar-based object detection, examining how varying environmental conditions, such as different weather patterns and road types, impact 3D object detection performance. Our findings highlight distinct domain shifts across various weather scenarios, revealing unique dataset sensitivities that underscore the critical role of radar point cloud generation. Additionally, we demonstrate that transitioning between different road types, especially from highways to urban settings, introduces notable domain shifts, emphasizing the necessity for diverse data collection across varied road environments. To the best of our knowledge, this is the first comprehensive analysis of domain shift effects on 4D radar-based object detection. We believe this empirical study contributes to understanding the complex nature of domain shifts in radar data and suggests paths forward for data collection strategy in the face of environmental variability.

Figures (5)

• Figure 1: Bird's eye view of radar object detection in heavy snow conditions from K-Radar dataset paek2022k: The boxes are ground truth, boxes are predictions from the model trained on all-weather data, and boxes are predictions from the model trained only on normal weather conditions.
• Figure 2: Illustration of the experimental workflow highlighting domain shift analysis between source and target datasets.
• Figure 3: Average power of received radar signal for different weather conditions. Notations: L. Snow represents Light Snow, and H. Snow denotes Heavy Snow.
• Figure 4: Evaluation of $AP_{BEV}$ and $AP_{3D}$ performance metrics for weather domain shift on Bosch-Radar data with 60k training samples (source$\rightarrow$target). N.:Normal, R.:Rain, O.:Overcast, M.:Mixed.
• Figure 5: $AP_{BEV}$ under normal and rainy conditions across varying training data sizes.