3D Small Object Detection Algorithm Based on Dynamic Feature Enhancement

doi:10.19678/j.issn.1000-3428.0252879

Abstract

Abstract:

In 3D object detection from point clouds, the inherent sparsity of Light Detection And Ranging (LiDAR) data poses pronounced challenges for small objects. A small number of effective points lead to weak structural cues and blurry boundaries; limited contextual awareness hinders spatial reasoning and semantic completion, causing localization bias; and the difficulty of precise spatial localization, weak channel expressiveness, and background dominance constrain accuracy. To mitigate the impact of the aforementioned issues on detection accuracy, a dynamic-aware 3D detector is proposed that integrates dynamic feature extraction with feature-enhancement mapping, targeting two critical stages of small-object detection: feature extraction and candidate generation. Specifically, a Dynamic Point Feature Prediction Network (DPFPN) that adaptively predicts and supplements sampling points to strengthen structural perception of small objects is introduced. Subsequently, a Feature Enhancement Mapping Network (FEMN) is built that deeply fuses the original features with those produced by the dynamic module to yield context-rich 2D feature maps, thereby compensating for contextual deficiency and improving localization. Finally, a Point Cloud Feature Enhancement Network (PCFEN) module is designed to sharpen focus on key small-object regions along both channel and spatial dimensions. Experiments on the nuScenes dataset demonstrate that the proposed approach performs better than mainstream detectors. Relative to the CenterPoint baseline, the mean Average Precision (mAP) increases from 56.1% to 59.4% and the Nuscenes Detection Score (NDS) rises from 64.4 to 67.4.

Key words: Light Detection And Ranging (LiDAR), dynamic feature extraction, feature enhancement, multi-scale features, object detection

摘要：

在点云三维目标检测任务中, 点云数据的稀疏性在客观上对小目标检测构成显著挑战。小目标自身有效点数稀少导致结构信息缺失与边界模糊, 而上下文感知能力不足阻碍模型有效利用周围环境信息进行空间推理与语义补全, 进而引发定位偏差, 以及其固有的空间定位困难、通道表达弱和特征易被背景淹没等问题, 共同制约了检测性能的提升。为缓解上述问题对检测精度造成的影响, 提出一种融合动态特征提取与特征增强映射的动态感知三维检测算法。该算法聚焦特征提取与候选框生成两大关键阶段对小目标检测进行优化。首先, 引入动态点特征预测网络(DPFPN), 通过自适应预测补充采样点以强化对小目标的结构感知能力; 其次, 构建特征增强映射网络(FEMN), 对原始特征及动态预测网络生成的特征进行深度融合, 输出富含上下文信息的二维特征图, 以有效弥补上下文缺失并提升小目标的定位精度; 最后, 设计点云特征增强网络(PCFEN), 在通道与空间双维度提升网络对小目标关键区域的聚焦能力。在nuScenes数据集上的实验结果表明, 所提算法性能优于当前主流目标检测算法, 与基准模型CenterPoint相比, 均值平均精度(mAP)由56.1%提升至59.4%, Nuscenes检测分数(NDS)由64.4提升至67.4。

关键词: 激光雷达, 动态特征提取, 特征增强, 多尺度特征, 目标检测

LI Luyang, YAN Jinlong, FANG Zeru, JIN Qiqi, XUE Hongxin. 3D Small Object Detection Algorithm Based on Dynamic Feature Enhancement[J]. Computer Engineering, 2026, 52(4): 264-275.

李潞洋, 闫锦龙, 方泽儒, 金旗旗, 薛红新. 基于动态特征增强的三维小目标检测算法[J]. 计算机工程, 2026, 52(4): 264-275.

/ Recommend / Download Citations

URL: https://www.ecice06.com/EN/10.19678/j.issn.1000-3428.0252879

https://www.ecice06.com/EN/Y2026/V52/I4/264

Figures/Tables 14

Fig.1 Comparison between the proposed method and traditional method

Fig.2 Overall framework of the proposed method

Fig.3 The characteristic flow direction among each module

Fig.4 Dynamic point feature prediction network

Fig.5 Feature enhancement mapping network

Fig.6 Point cloud feature enhancement network

Fig.7 Visualization of point cloud feature enhancement network

References 29

1	李昌财, 陈刚, 侯作勋, 等. 自动驾驶中的三维目标检测算法研究综述. 中国图象图形学报, 2024, 29 (11): 3238- 3264.
	LI C C , CHEN G , HOU Z X , et al. A review of research on three-dimensional target detection algorithms in autonomous driving. Chinese Journal of Image Graphics, 2024, 29 (11): 3238- 3264.
2	ZHANG B , WANG H S , YOU S L , et al. A small-size 3Dobject detection network for analyzing the sparsity of raw LiDAR point cloud. Journal of Russian Laser Research, 2023, 44 (6): 646- 655. doi: 10.1007/s10946-023-10173-3
3	WANG J , LIU Y , ZHU Y , et al. 3D point cloud object detection method based on multi-scale dynamic sparse voxelization. Sensors, 2024, 24 (6): 1804. doi: 10.3390/s24061804
4	YANG H H, WANG W X, CHEN M H, et al. PVT-SSD: single-stage 3D object detector with point-voxel transformer[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington D. C., USA: IEEE Press, 2023: 13476-13487.
5	龙丽叶, 焦世超, 郭磊, 等. 基于紧凑中心的多模态三维模型检索研究. 计算机工程, 2025, 51 (2): 322- 334. doi: 10.19678/j.issn.1000-3428.0069685
	LONG L Y , JIAO S C , GUO L , et al. Research on multimodal 3D model retrieval based on compact center. Computer Engineering, 2025, 51 (2): 322- 334. doi: 10.19678/j.issn.1000-3428.0069685
6	CHARLES R Q, HAO S, MO K C, et al. PointNet: deep learning on point sets for 3D classification and segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Washington D. C., USA: IEEE Press, 2017: 652-660.
7	QI C R, YI L, SU H, et al. Pointnet++: deep hierarchical feature learning on point sets in a metric space[C]//Proceedings of the Advances in Neural Information Processing Systems. Cambridge, USA: MIT Press, 2017: 301-312.
8	SHI S S, WANG X G, LI H S. PointRCNN: 3D object proposal generation and detection from point cloud[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach, USA: IEEE Press, 2019: 770-779.
9	ZHOU Y, TUZEL O. VoxelNet: end-to-end learning for point cloud based 3D object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE Press, 2018: 4490-4499.
10	LANG A H, VORA S, CAESAR H, et al. PointPillars: fast encoders for object detection from point clouds[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington D. C., USA: IEEE Press, 2019: 12697-12705.
11	YAN Y , MAO Y , LI B . SECOND: sparsely embedded convolutional detection. Sensors, 2018, 18 (10)
12	DENG J J , SHI S S , LI P W , et al. Voxel R-CNN: towards high performance voxel-based 3D object detection. Artificial Intelligence, 2021, 35 (2): 1201- 1209.
13	YIN T, ZHOU X, KRAHENBUHL P. Center-based 3D object detection and tracking[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington D. C., USA: IEEE Press, 2021: 11784-11793.
14	孙一杰, 李晓明. 结合动态循环金字塔与任务解耦的无锚框检测. 计算机工程与设计, 2025, 46 (4): 1157- 1166.
	SUN Y J , LI X M . Combining dynamic cyclic pyramid with task decoupling for anchor-free frame detection. Computer Engineering and Design, 2025, 46 (4): 1157- 1166.
15	LAW H, DENG J. CornerNet: detecting objects as paired keypoints[C]// Proceedings of the European conference on Computer Vision. Berlin, Germany: Springer, 2018: 765-781.
16	ZHOU X Y, ZHUO J C, KRAHENBUHL P. Bottom-up object detection by grouping extreme and center points[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach, USA: IEEE Press, 2019: 850-859.
17	ZHOU X, WANG D, KRAHENBUHL P. Objects as points[EB/OL]. [2025-07-10]. https://arxiv.org/abs/1904.07850.
18	TIAN Z, SHEN C H, CHEN H, et al. FCOS: fully convolutional one-stage object detection[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. Washington D. C., USA: IEEE Press, 2019: 9627-9636.
19	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. Seattle, USA: IEEE Press, 2020: 11621-11631.
20	LOSHCHILOV I, HUTTER F. Decoupled weight decay regularization[EB/OL]. [2025-07-10]. https://arxiv.org/abs/1711.05101.
21	HU P Y, ZIGLAR J, HELD D, et al. What you see is what you get: exploiting visibility for 3D object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, USA: IEEE Press, 2020: 11001-11009.
22	CHEN Q, SUN L, CHEUNG E, et al. Every view counts: cross-view consistency in 3d object detection with hybrid-cylindrical-spherical voxelization[C]//Proceedings of the Advances in Neural Information Processing Systems. Cambridge, USA: MIT Press, 2020: 21224-21235.
23	VORA S, LANG A H, HELOU B, et al. PointPainting: sequential fusion for 3D object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, USA: IEEE Perss, 2020: 4604-4612.
24	YIN J B, SHEN J B, GUAN C Y, et al. LiDAR-based online 3D video object detection with graph-based message passing and spatiotemporal transformer attention[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington D. C., USA: IEEE Press, 2020: 11495-11504.
25	ZHU X G, MA Y X, WANG T, et al. SSN: shape signature networks for multi-class object detection from point clouds[C]//Proceedings of the European conference on Computer Vision. Berlin, Germany: Springer, 2020: 581-597.
26	ZHU B, JIANG Z, ZHOU X, et al. Class-balanced grouping and sampling for point cloud 3d object detection[EB/OL]. [2025-07-10]. https://arxiv.org/abs/1908.09492.
27	CHEN C, CHEN Z, ZHANG J, et al. SASA: semantics-augmented set abstraction for point-based 3d object detection[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Washington D. C., USA: IEEE Press, 2022: 221-229.
28	YANG Z, SUN Y, LIU S, et al. 3DSSD: point-based 3d single stage object detector[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington D. C., USA: IEEE Press, 2020: 11040-11048.
29	YANG H, WANG W, CHEN M, et al. PVT-SSD: single-stage 3d object detector with point-voxel transformer[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Washington D. C., USA: IEEE Press, 2023: 13476-13487.

[1]	HAO Yousheng, WEN Zhenhui, FENG Xiaoxi, DENG Zehua, HUANG Qingbao. Vehicle Paint Defect Detection Based on Improved YOLOv8 [J]. Computer Engineering, 2026, 52(4): 252-263.
[2]	LI Yiyang, LU Shenglian, WANG Jijie, CHEN Ming. Review of DETR Object Detection Algorithm Based on Transformer [J]. Computer Engineering, 2026, 52(4): 62-81.
[3]	CAO Jiwei, LUO Fei, DING Weichao. BS-YOLO: A Small Object Detection Algorithm Based on BSAM Attention Mechanism and SCConv [J]. Computer Engineering, 2026, 52(3): 119-127.
[4]	XIE Binhong, SHI Yufei, ZHANG Rui, ZHANG Yingjun. Few-shot Object Detection Method Based on Query Guidance and Semantic Enhancement [J]. Computer Engineering, 2026, 52(3): 141-151.
[5]	WEN Lang, GOU Guanglei, BAI Ruifeng, MIAO Wanyu. Few-shot Fine-grained Image Classification Based on Neighborhood Fusion and Feature Enhancement [J]. Computer Engineering, 2026, 52(2): 158-166.
[6]	LI Jianlang, WU Xindian, CHEN Ling, YANG Bo, TANG Wensheng. 3D Object Detection Algorithm Based on 4D Millimeter-Wave Radar and Vision Fusion [J]. Computer Engineering, 2026, 52(2): 299-310.
[7]	LI Shugang, LI Shuang, LIU Chi. Cross-Domain Fabric Defect Detection Guided by Texture Knowledge [J]. Computer Engineering, 2026, 52(1): 166-175.
[8]	MA Gan, GU Yu, PENG Dongliang. Combining Improved YOLOv5s and Dynamic Data Augmentation for Sea Surface Ship Detection [J]. Computer Engineering, 2025, 51(9): 294-305.
[9]	WANG Shumeng, XU Huiying, ZHU Xinzhong, HUANG Xiao, SONG Jie, LI Yi. Lightweight Small Object Detection Algorithm for Aerial Photography Based on Improved YOLOv8n: PECS-YOLO [J]. Computer Engineering, 2025, 51(9): 280-293.
[10]	ZHU Siyuan, LI Jiasheng, ZOU Danping, HE Di, YU Wenxian. Unstructured Road Defect Detection Algorithm Based on Semi-Supervised Learning [J]. Computer Engineering, 2025, 51(9): 14-24.
[11]	MIAO Ru, LI Yi, ZHOU Ke, ZHANG Yanna, CHANG Ranran, MENG Geng. A Study on Improved Faster R-CNN Model for Multi-Object Detection in Remote Sensing Images [J]. Computer Engineering, 2025, 51(8): 292-304.
[12]	SONG Jie, XU Huiying, ZHU Xinzhong, HUANG Xiao, CHEN Chen, WANG Zeyu. Improved Fall Detection Algorithm Based on YOLOv8: OEF-YOLO [J]. Computer Engineering, 2025, 51(7): 127-139.
[13]	LIU Xudong, YANG Xubing. Design of L1-OCSVM Model and Its Application in Forestry Object Detection [J]. Computer Engineering, 2025, 51(7): 375-384.
[14]	ZHANG Jiacheng, WEI Jin, CHEN Yishi. Improved YOLOv8 Real-time Lightweight Robust Hedge Detection Algorithm [J]. Computer Engineering, 2025, 51(7): 362-374.
[15]	LUAN Mengna, ZHENG Qiumei, WANG Fenghua. Real-time Traffic Sign Detection Algorithm Based on DMC-YOLO [J]. Computer Engineering, 2025, 51(7): 90-99.

Please choose a citation manager

Content to export