MotionNet-PGA: MotionNet with Polar-Guided Attention for Moving Object Segmentation in Scanning Radar
RenYi Yuan, Chieh-Chih Wang, Wen-Chieh Lin
AI summary
Problem
Scanning radar’s low angular resolution, spatial distortion, and clutter make accurate moving object segmentation extremely difficult, particularly for small or distant objects, while existing LiDAR and camera methods fail to transfer to this modality.
Approach
MotionNet-PGA adapts a spatio-temporal convolution backbone for radar and introduces a polar-guided attention module that fuses Cartesian motion cues with polar-domain appearance features to suppress clutter and preserve geometric fidelity.
Key results
- Proposes MotionNet-PGA with a polar-guided attention module
- Introduces the first annotated scanning radar MOS dataset
- Achieves 68.36% overall IoU, surpassing MotionNet by 2.48%
- Delivers 4.08% improvement for small object segmentation
Why it matters
Provides a reliable, radar-native perception solution for autonomous driving and robotics, enabling robust dynamic scene understanding without relying on LiDAR or cameras.
Abstract
Moving object segmentation (MOS) is essential for autonomous driving, enabling robust detection, tracking, and prediction of dynamic agents in complex traffic scenarios. Radar sensors offer notable advantages for long-range sensing, but their lower spatial resolution, measurement noise, and geometric distortions—particularly for distant targets—pose significant challenges for accurate MOS. These limitations are amplified when detecting small objects such as scooters. In this work, we present MotionNet-PGA, a Polar-guided Attention Framework designed specifically for scanning radar- based MOS. Our method builds on the multi-frame motion encoding backbone of MotionNet [1], and introduces a polar- guided attention module to suppress clutter, enhance motion feature representation, and improve segmentation of small and distant targets. For evaluation, we construct and annotate the ITRI Radar moving object segmentation Dataset. Experimental results demonstrate that our method surpasses state-of-the-art baseline, MotionNet, by 2.48% in overall IoU and achieves a 4.08% improvement in small-object segmentation. These results highlight the effectiveness of polar-guided attention in addressing scanning radar-specific challenges.