Accuracy- and Real-Time-Aware 4D Radar Preprocessing for Autonomous Driving Perception Systems

Woo-Jin Jung; Dong-Hee Paek; Jeong-Su Park; Seung-Hyun Kong

cs.CVSep 16, 2026

Accuracy- and Real-Time-Aware 4D Radar Preprocessing for Autonomous Driving Perception Systems

Authors: Woo-Jin Jung, Dong-Hee Paek, Jeong-Su Park, Seung-Hyun Kong

Organizations: Cho Chun Shik Graduate School of Mobility Korea Advanced Institute of Science and Technology (KAIST) Daejeon 34051, Republic of Korea · Mechanical Engineering Research Institute Korea Advanced Institute of Science and Technology (KAIST) Daejeon 34141, Republic of Korea · Autonomous Driving Perception Technology Vanguard Team Hyundai Motor Company Seongnam-si 13529, Republic of Korea

Abstract

4D radar has emerged as a promising next-generation sensor for improving the robustness of autonomous driving perception systems because of its stable sensing capability under adverse weather conditions. However, deploying 4D radar in embedded environments with limited hardware resources requires radar-representation preprocessing that jointly considers perception accuracy, real-time performance, and computational complexity. This paper proposes a preprocessing framework for 4D-radar-based 3D object detection. First, Percentile-based 3D Shape Preservation (P3DP) extracts point clouds from radar tensors while preserving object-shape information and suppressing noise and false alarms. Second, Multi-frame-based Noise Point Discrimination using Kernel Density Estimation (MF-KDE) improves the density and reliability of sparse radar point clouds. Finally, Embedded & NetScore (ENS) evaluates suitability for embedded deployment by jointly considering accuracy, real-time performance, adverse-weather robustness, and model complexity.

Figures & tables

Fig. 1: Data representations of 4D radar. A tensor provides richer spatial information but is computationally expensive, whereas a point cloud is lightweight but sparse and less effective for representing object shape.

Fig. 2: Overview of the 4D radar perception system. ADC denotes analog-to-digital converter. Tensor preprocessing and point-cloud preprocessing generate radar representations that are subsequently used as inputs to the DNN-based object-detection network.

Panel of Figure 4 — Fig. 4: (a) Tensor power-density estimation and (b) visualization of the power density. Black points denote LiDAR measurements, and colored points denote percentile-extracted radar points. Red indicates high power in (a), whereas it indicates low power density in (b) to make object boundaries easier to distinguish.

Fig. 6: Visualization of point-density augmentation: (a) multi-frame alignment and (b) the result after adding the KDE-based point-density feature. Color represents the normalized density value (0: blue; 1: red). Points around objects exhibit relatively high density, whereas isolated points exhibit low density, providing an auxiliary feature for distinguishing object points from noise.

Detection model	Preprocessing	$\mathbf{BEV_{AP}}$ [%]	$\mathbf{3D_{AP}}$ [%]	FPS [Hz]	ENS [%]
RTNH [ 1 ]	CA-CFAR	64.42	55.47	22.87	52.64
	Percentile 5%	64.89	56.05	22.65	53.84
	P3DP	66.58	59.99	25.60	54.21
RadarPillar-Net [ 15 ]	CA-CFAR	63.67	54.12	36.45	64.18
	Percentile 5%	57.63	54.10	36.56	63.49
	P3DP	65.86	55.69	27.29	64.43

TABLE I: Results of tensor preprocessing.

Fig. 7: Comparison of tensor-preprocessing results.

Detection model	Preprocessing	$\mathbf{BEV_{AP}}$ [%]	$\mathbf{3D_{AP}}$ [%]	FPS [Hz]	ENS [%]
PointPillars [ 26 ]	Not applied	33.30	31.05	88.09	48.55
	Alignment	33.60	32.57	33.89	48.76
	DoppDrive	33.65	32.64	31.28	49.02
	MF-KDE	33.68	32.72	29.57	49.35
RadarPillar-Net [ 15 ]	Not applied	34.14	33.63	63.48	49.00
	Alignment	34.29	33.85	28.56	49.08

TABLE II: Results of point-cloud preprocessing.

Autonomous Driving Perception

Autonomous driving perception enables vehicles to detect and interpret road users, drivable space, and hazards using onboard and shared sensors.

See all 379 papers

STAR: Scene- and Task-Aware 4D Radar Preprocessing Towards End-to-End Cognitive Radar
Sep 21, 2026Seung-Hyun Song, Dong-Hee Paek, Seung-Hyun KongAutonomous Driving Perception 3D Object Detection
Cho Chun Shik Graduate School of Mobility, Korea Advanced Institute of Science and Technology (KAIST), Daejeon 34141, Republic of Korea · Department of Future Mobility, Korea University, Sejong 30019, Republic of Korea
RCGDet3D: Rethinking 4D Radar-Camera Fusion-based 3D Object Detection with Enhanced Radar Feature Encoding
May 20, 2026Weiyi Xiong, Bing ZhuCamera-Radar Fusion 3D Object Detection
School of Automation Science and Electrical Engineering, Beihang University, Beijing, P.R. China
Sparse4D-Radar: An Efficient and Robust Framework for Surround-View 3D Object Detection via 4D Radar-Camera Fusion
Jul 5, 2026Fuyuan Ai, Yuchen Tan, Jiehui Chen +2Camera-Radar Fusion Autonomous Driving Perception
State Key Laboratory of Ocean Sensing and Ocean College, Zhejiang University, Zhoushan 316021, China, and also with the Engineering Research Center of Oceanic Sensing Technology and Equipment, Ministry of Education, Zhejiang University, Zhoushan 316021, China · Donghai Laboratory, Zhoushan 316021, China, also with the Ocean College, Institute of Marine Electronics and Intelligent Systems, Zhejiang University, Zhoushan 316021, China, and also with the Engineering Research Center of Oceanic Sensing Technology and Equipment, Ministry of Education, Zhejiang University, Zhoushan 316021, China

Sep 21, 2026J/K move · Enter open · S save

STAR: Scene- and Task-Aware 4D Radar Preprocessing Towards End-to-End Cognitive Radar

Seung-Hyun Song, Dong-Hee Paek, Seung-Hyun Kong

Cho Chun Shik Graduate School of Mobility, Korea Advanced Institute of Science and Technology (KAIST), Daejeon 34141, Republic of Korea · Department of Future Mobility, Korea University, Sejong 30019, Republic of Korea

Autonomous Driving Perception 3D Object Detection Object Detection

Open paper AskPDF arXiv

Four-dimensional (4D) Radar has emerged as a key sensor for environmental perception, providing range, azimuth, elevation, and Doppler measurements while remaining robust to illumination changes and adverse weather conditions. However, conventional Radar preprocessing methods, such as constant false alarm rate (CFAR) detection, select measurements primarily based on signal-level criteria and may therefore discard information valuable for downstream perception during point cloud generation. In addition, existing 4D Radar perception pipelines typically optimize Radar data processing and downstream perception independently, preventing task objectives from directly guiding the preprocessing stage. To address these limitations, we propose a Scene- and Task-Aware Radar (STAR) Preprocessor together with an end-to-end training framework. The STAR Preprocessor incorporates scene context and downstream task objectives to generate task-relevant Radar points, enabling the Radar representation to be optimized directly for perception. On the K-Radar benchmark, the proposed method achieves 74.3 AP, outperforming the previous state of the art by 5.6 AP points. Furthermore, applying the task-relevant points generated by STAR to various existing 3D detectors improves detection performance in most evaluation settings and yields an overall positive average gain over point clouds produced by conventional preprocessing.

Abstract

Autonomous Driving Perception

Explore similar work

STAR: Scene- and Task-Aware 4D Radar Preprocessing Towards End-to-End Cognitive Radar

RCGDet3D: Rethinking 4D Radar-Camera Fusion-based 3D Object Detection with Enhanced Radar Feature Encoding

Sparse4D-Radar: An Efficient and Robust Framework for Surround-View 3D Object Detection via 4D Radar-Camera Fusion