基于极坐标的环视视觉稀疏化时序3D目标检测

doi:10.19562/j.chinasae.qcgc.2025.06.018

Abstract

Abstract:

To address the trade-off between accuracy and real-time performance in vision-based surround-view 3D object detection for autonomous vehicles， PolarSparse4D， a sparse query-based method using polar parametrization， is proposed. The model consists of an image encoder， a 3D anchor decoder and an auxiliary quality assessment branch for training. Firstly， to avoid the detection distance limitation caused by parameter normalization， a feature encoding method that decouples the center distance and azimuth angle of the 3D anchor boxes is designed. Secondly， by designing an anchor spatial information interaction self-attention module and a temporal information fusion module， the spatiotemporal information fusion process of anchors is completed efficiently and accurately. Finally， an anchor box parameter quality assessment branch is established to improve the detection accuracy and model convergence speed significantly. The experimental results on the nuScenes validation set show that the proposed model achieves 41.3% and 52.5% on mAP and NDS， respectively， with a speed of 19.2 FPS， demonstrating high accuracy and real-time capability.

Key words: 3D object detection, surround-view camera, polar parametrization, autonomous vehicle

Chao Wei,Shuxin Sui,Luxing Li. PolarSparse4D: Polar Parametrization for Vision-Based Surround-View Temporal Sparse 3D Object Detection[J].Automotive Engineering, 2025, 47(6): 1198-1206.

Figures/Tables 11

$λ 1$	$λ 2$	$λ 3$	$ξ 1$	$ξ 2$	$ξ 3$
2.0	0.25	1.0	1.0	1.0	1.0

References 27

1	PHILION J， FIDLER S. Lift， splat， shoot： encoding images from arbitrary camera rigs by implicitly unprojecting to 3D［C］. Computer Vision-ECCV 2020： 16th European Conference， Glasgow， UK， August 23-28， 2020， Proceedings， Part XIV 16. Springer International Publishing， 2020： 194-210.
2	HUANG J， HUANG G， ZHU Z， et al. BEVDet： high-performance multi-camera 3D object detection in bird-eye-view［J］. arXiv preprint arXiv：， 2021.
3	HUANG J， HUANG G. BEVDet4D： exploit temporal cues in multi-camera 3D object detection［J］. arXiv preprint arXiv：， 2022.
4	XIE E， YU Z， ZHOU D， et al. M² BEV： multi-camera joint 3D detection and segmentation with unified birds-eye view representation［J］. arXiv preprint arXiv：， 2022.
5	HUANG J， HUANG G. BEVPoolv2： a cutting-edge implementation of BEVDet toward deployment［J］. arXiv preprint arXiv：， 2022.
6	LI Y， HUANG B， CHEN Z， et al. Fast-BEV： a fast and strong bird's-eye view perception baseline［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence， 2024.
7	WANG Y， GUIZILINI V C， ZHANG T， et al. DETR3D： 3D object detection from multi-view images via 3D-to-2D queries［C］. Conference on Robot Learning. PMLR， 2022： 180-191.
8	LI Z， WANG W， LI H， et al. BEVFormer： learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers［C］. European Conference on Computer Vision. Cham： Springer Nature Switzerland， 2022： 1-18.
9	YANG C， CHEN Y， TIAN H， et al. BEVFormer v2： adapting modern image backbones to bird's-eye-view recognition via perspective supervision［C］. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition， 2023： 17830-17839.
10	LIN X， LIN T， PEI Z， et al. Sparse4D： multi-view 3D object detection with sparse spatial-temporal fusion［J］. arXiv preprint arXiv：， 2022.
11	LIN X， LIN T， PEI Z， et al. Sparse4D v2： recurrent temporal fusion with sparse model［J］. arXiv preprint arXiv：， 2023.
12	LIN X， PEI Z， LIN T， et al. Sparse4D v3： advancing end-to-end 3D detection and tracking［J］. arXiv preprint arXiv：， 2023.
13	LIU H， TENG Y， LU T， et al. SparseBEV： high-performance sparse 3D object detection from multi-camera videos［C］. Proceedings of the IEEE/CVF International Conference on Computer Vision， 2023： 18580-18590.
14	JIANG Y， ZHANG L， MIAO Z， et al. PolarFormer： multi-camera 3D object detection with polar transformer［C］. Proceedings of the AAAI conference on Artificial Intelligence， 2023， 37（1）： 1042-1050.
15	CHEN S， WANG X， CHENG T， et al. Polar parametrization for vision-based surround-view 3D detection［J］. arXiv preprint arXiv：， 2022.
16	VASWANI A. Attention is all you need［J］. Advances in Neural Information Processing Systems， 2017.
17	KUHN H W. The Hungarian method for the assignment problem［J］. Naval Research Logistics Quarterly， 1955， 2（1‐2）： 83-97.
18	LIN T. Focal loss for dense object detection［J］. arXiv preprint arXiv：， 2017.
19	WANG J， LI F， BI H. Gaussian focal loss： learning distribution polarized angle prediction for rotated object detection in aerial images［J］. IEEE Transactions on Geoscience and Remote Sensing， 2022， 60： 1-13.
20	CAESAR H， BANKITI V， LANG A H， et al. nuScenes： a multimodal dataset for autonomous driving［C］. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition， 2020： 11621-11631.
21	PASZKE A， GROSS S， MASSA F， et al. Pytorch： an imperative style， high-performance deep learning library［J］. Advances in Neural Information Processing Systems， 2019， 32.
22	HE K， ZHANG X， REN S， et al. Deep residual learning for image recognition［C］. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition， 2016： 770-778.
23	LEE Y， PARK J. CenterMask： real-time anchor-free instance segmentation［C］. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition， 2020： 13906-13915.
24	LOSHCHILOV I. Decoupled weight decay regularization［J］. arXiv preprint arXiv：， 2017.
25	LIU Y， YAN J， JIA F， et al. Petrv2： a unified framework for 3D perception from multi-camera images［C］. Proceedings of the IEEE/CVF International Conference on Computer Vision， 2023： 3262-3272.
26	LI Y， BAO H， GE Z， et al. Bevstereo： enhancing depth estimation in multi-view 3D object detection with temporal stereo［C］. Proceedings of the AAAI Conference on Artificial Intelligence， 2023， 37（2）： 1486-1494.
27	HUANG J， HUANG G. Bevpoolv2： a cutting-edge implementation of bevdet toward deployment［J］. arXiv preprint arXiv：， 2022.

坐标系	方法	主干	图像尺寸	NDS↑	mAP↑	mATE↓	mASE↓	mAOE↓	mAVE↓	mAAE↓	FPS↑
笛卡尔坐标系	DETR3D^［7］	R50	900×1600	37.3%	30.2%	0.811	0.282	0.493	0.979	0.212	6.3
	BEVDet^［2］	R50	256×704	37.2%	28.6%	0.724	0.278	0.590	0.873	0.247	7.8
	BEVDet^［2］	R50	384×1056	38.1%	30.4%	0.719	0.272	0.555	0.903	0.257	4.2
	PETRv2^［25］	R50	256×704	45.6%	34.9%	0.700	0.275	0.580	0.437	0.187
	FastBEV^［6］	R50	256×704	48.7%	35.4%	0.656	0.281	0.384	0.361	0.217	41.3
	BEVStereo^［26］	R50	256×704	50.0%	37.2%	0.598	0.270	0.438	0.367	0.190
	BEVPoolv2^［27］	R50	256×704	52.6%	40.6%	0.572	0.275	0.463	0.275	0.188	16.6
	BEVFormerV2^［9］	R50		52.9%	42.3%	0.618	0.273	0.413	0.333	0.188
	Sparse4Dv2^§［11］	R50	256×704	52.1%	41.0%	0.609	0.272	0.449	0.330	0.187	19.9
极坐标系	PolarDETR-T^［15］	R50	900×1600	45.8%	35.4%	0.748	0.277	0.432	0.539	0.197	~6.0
	PolarDETR-T^［15］	R101	900×1600	48.8%	38.3%	0.707	0.269	0.344	0.518	0.196	~3.5
	PolarFormer^［14］	R101	900×1600	47.0%	41.5%	0.657	0.263	0.405	0.911	0.139
	PolarSparse4D	R50	256×704	52.5%	41.3%	0.599	0.270	0.508	0.256	0.185	19.2

类别	AP↑	ATE↓	ASE↓	AOE↓	AVE↓	AAE↓
轿车	0.616	0.395	0.144	0.070	0.201	0.196
货车	0.314	0.607	0.202	0.070	0.219	0.212
公共汽车	0.403	0.740	0.194	0.086	0.440	0.212
拖车	0.144	1.006	0.248	0.928	0.174	0.053
建筑工程车	0.099	1.002	0.502	1.202	0.130	0.436
行人	0.489	0.575	0.291	0.594	0.339	0.175
摩托车	0.421	0.523	0.257	0.690	0.372	0.227
自行车	0.398	0.455	0.249	1.115	0.147	0.008
交通锥	0.667	0.305	0.316
路障	0.575	0.365	0.286	0.146

解码器	NDS↑	mAP↑	GFlops↓	Params↓
NT+5T	52.3%	41.1%	167.995	47.905 M
NT+4T+NT	52.4%	41.1%	167.798	46.854 M
NT+3T+2NT	52.3%	41.0%	167.602	45.804 M
NT+2T+3NT	52.5%	41.2%	167.405	44.753 M
NT+1T+4NT	52.5%	41.3%	167.209	43.702 M

C	Y	Az	NDS↑	mAP↑	GFlops↓	Params↓
			51.8%	40.8%	167.207	43.697 M
√			52.0%	40.9%	167.208	43.699 M
√	√		52.3%	41.0%	167.208	43.700 M
√	√	√	52.5%	41.3%	167.209	43.702 M

[1]	Lingyun Zhu,Haiyang Wang. Autonomous Vehicle Object Detection by LiDAR Point Cloud Feature Completion in Snowfall Scenarios [J]. Automotive Engineering, 2025, 47(6): 1133-1143.
[2]	Hai Wang,Guirong Zhang,Tong Luo,Meng Qiu,Yingfeng Cai,Long Chen. A Multi-modal Data Mining Algorithm for Corner Case of Automatic Driving Road Scene [J]. Automotive Engineering, 2024, 46(7): 1239-1248.
[3]	Guodong Wang,Li Liu,Yu Meng,Haiping Du,Guoxing Bai,Qing Gu. Research on Collision Avoidance Limit of Autonomous Vehicles [J]. Automotive Engineering, 2024, 46(6): 985-994.
[4]	Guodong Du,Yuan Zou,Xudong Zhang,Wenjing Sun,Wei Sun. Research on Automatic Driving Motion Control Based on Double Estimator Reinforcement Learning Combined with Forward Predictive Control [J]. Automotive Engineering, 2024, 46(4): 564-576.
[5]	Xinke Fu,Yingfeng Cai,Long Chen,Hai Wang,Qingchao Liu. Decision-Making for Autonomous Driving in Uncertain Environment [J]. Automotive Engineering, 2024, 46(2): 211-221.
[6]	Mengxi Wang,Yingfeng Cai,Hai Wang,Zhongyu Rao,Long Chen,Yicheng Li. Vehicle Trajectory Prediction Method Based on Graph Convolutional Interaction Network [J]. Automotive Engineering, 2024, 46(10): 1863-1872.
[7]	Cheng Lin, Bowen Wang, Lü Peiyuan, Xinle Gong, Xiao Yu. Research on Motion Planning and Cooperative Control for Autonomous Vehicles with Lane Change Gaming Maneuvers Under the Curved Road [J]. Automotive Engineering, 2023, 45(7): 1099-1111.
[8]	Dongyu Zhao, Shuen Zhao. Autonomous Driving 3D Object Detection Based on Cascade YOLOv7 [J]. Automotive Engineering, 2023, 45(7): 1112-1122.
[9]	Yanyan Chen,Hai Wang,Yingfeng Cai,Long Chen,Yicheng Li. Efficient Automatic Driving Instance Segmentation Method Based on Detection [J]. Automotive Engineering, 2023, 45(4): 541-550.
[10]	Ziwei Zhang,Ling Zheng,Yinong Li,Xuqiang Qiao,Hao Zheng,Kan Wang. A Multi-objective Adaptive Cruise Control Strategy for Autonomous Vehicle Considering Uncertain Movements of Preceding Vehicle [J]. Automotive Engineering, 2023, 45(3): 361-371.
[11]	Xinrong Zhang,Yuhang Tan,Yifan Jia,Jin Huang,Quanning Xu. Robust Control of Path Tracking for Four-Wheel Independent Drive Electric Vehicles [J]. Automotive Engineering, 2023, 45(2): 253-262.
[12]	Lü Ying,Xu Qi,Qiuzheng Liu,Xinyu Wang,Guoying Chen. Path Tracking Control Method with Steering Lag for Autonomous Vehicles [J]. Automotive Engineering, 2023, 45(12): 2234-2241.
[13]	Zhengfa Liu,Ya Wu,Peigen Liu,Rongqi Gu,Guang Chen. Cross-Domain Object Detection for Intelligent Driving Based on Joint Distribution Matching of Features and Labels [J]. Automotive Engineering, 2023, 45(11): 2082-2091.
[14]	Long Chen,Chen Yang,Yingfeng Cai,Hai Wang,Yicheng Li. Pedestrian Crossing Intention Prediction Method Based on Multimodal Feature Fusion [J]. Automotive Engineering, 2023, 45(10): 1779-1790.
[15]	Fengchong Lan,Yingjie Liu,Jiqing Chen,Zhaolin Liu. Study on Motion Planning of Autonomous Vehicles in Cut-in Scenes Based on Dynamic Game Algorithm [J]. Automotive Engineering, 2023, 45(1): 9-19.

PolarSparse4D: Polar Parametrization for Vision-Based Surround-View Temporal Sparse 3D Object Detection

RichHTML

PDF (PC)

Like

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 11

References 27

Related Articles 15

Metrics

Comments

Recommended 0