文献类型：期刊文献

英文题名：DDCFusion: Dynamic Depth Compensation Fusion for CameraRadar 3-D Object Detection

作者：Chen, Jiahao[1];Chen, Huanlei[2];Zhu, Ziming[1];Shen, Zheng[1];Ling, Xiaofeng[1,3];Zhu, Yu[1]

机构：[1]East China Univ Sci & Technol, Sch Informat Sci & Engn, Shanghai 200237, Peoples R China;[2]Shanghai Motor Vehicle Inspect Certificat & Tech I, Shanghai 201805, Peoples R China;[3]East China Univ Sci & Technol, Shanghai Key Lab Intelligent Sensing & DetectionTe, Shanghai 200237, Peoples R China

年份：2026

卷号：26

期号：3

起止页码：4561

外文期刊名：IEEE SENSORS JOURNAL

收录：;EI(收录号：20260119863754);WOS:【SCI-EXPANDED(收录号:WOS:001676275200040)】；

基金：This work was supported in part by the National Natural Science Foundation of China under Grant 62476088, in part by Shanghai Automotive Industry Science and Technology Development Foundation under Grant 2304, and in part by the Science and Technology Commission of Shanghai Municipality under Grant 20DZ2254400.

语种：英文

外文关键词：Radar; Radar cross-sections; Point cloud compression; Radar imaging; Radar detection; Feature extraction; Three-dimensional displays; Object detection; Doppler radar; Cameras; 3-D object detection; 4-D millimeter-wave (4D-MMW) radar; autonomous driving; camera; dynamic depth compensation (DDC); multimodal fusion

摘要：The effective representation and feature extraction from sparse point clouds of 4-D millimeter-wave (4D-MMW) radar pose a significant challenge in 3-D object detection. This article proposes dynamic depth compensation fusion (DDCFusion), a novel radar-camera fusion network that advances measurement precision by dynamically compensating for depth errors in sparse radar data. DDCFusion achieves this by exploiting the physical properties of 4-D MMW radar to improve measurement reliability and reduce depth uncertainty, which enhances depth measurement confidence in the view transform by integrating radar cross section (RCS)-derived reflectivity metrics. The occupancy (OCC)-weighted radar branch prioritizes image regions with high-confidence radar returns, minimizing the measurement noise in the view transform operation. Furthermore, DDCFusion optimizes spatial measurement consistency in bird's-eye-view (BEV) space by modeling cross-sensor dependencies through the global feature slice coordinate attention (GFSCA) fusion module. Experimental validation on the VoD and TJ4DRadSet datasets demonstrates superior measurement accuracy, achieving 51.08% mean average precision (mAP) on VoD and 34.61% mAP on TJ4DRadSet-outperforming existing methods in depth error reduction and robustness to sparsity. Ablation studies verify the measurement-centric design: RCS-guided diffusion improves small-object detection (e.g., pedestrians), while densitybased spatial clustering of applications with noise (DBSCAN)-based clustering refines large-object localization (e.g., vehicles). The network demonstrates significant improvements in depth accuracy and robustness to sparse inputs while maintaining the competitive inference latency with 138 ms.