DPRT-YOLO：智能网联汽车复杂驾驶环境实时目标检测器

doi:10.3778/j.issn.1002-8331.2412-0023

摘要/Abstract

摘要： 目标检测是智能网联汽车视觉感知系统的一项基本任务，可为先进驾驶辅助系统提供基础数据和决策依据。然而，在低光照和恶劣天气等复杂环境中，车载目标检测算法面临小目标检测性能不佳、漏检率和误检率偏高的挑战。针对这一挑战，发展了一种面向智能网联汽车的实时目标检测器（DPRT-YOLO），通过对流行的YOLOv10模型进行改造，使其更加适用于复杂驾驶环境中的目标检测任务，并通过在NVIDIA边缘计算平台上开展消融和对比实验，验证了算法的有效性。设计了增强加权多分支特征融合网络（EWMFFN），引入浅层加权融合和多分支加权融合模块，消除特征融合过程中的层间干扰，设计星形拓扑特征交互结构，提升模型对小尺度目标的检测能力，同时保持了网络结构的轻量化设计。融合卷积门控线性单元（convolutional gated linear units，CGLU）与卷积加法自注意力（convolutional additive token mixer，CATM），通过局部-全局双通路机制建立小目标尺度信息的长期上下文关系并保持模型的轻量化。为了评估模型在真实算力场景中的检测性能，将其部署在NVIDIA Jetson Xavier Nx平台上，采用NVIDIA TensorRT FP16量化加速，在BDD100K和TT100K测试集上开展推理实验，并与基准模型进行对比，结果显示：（1）检测精度方面，与YOLOv10n和YOLO11n相比，改进模型的mAP@0.5指标分别提升了6.1和7.4个百分点，mAP@0.5：0.95指标分别提升了3.6和4.2个百分点，同时，参数量分别降低了26.1%和34.9%。（2）检测速度方面，改进模型Small和Nano两种版本的推理速度分别达到了29?FPS和35?FPS。实验结果表明：与参考模型相比，改进算法在复杂驾驶环境中的表现更加优异，在检测精度与检测速度之间达到了更好的平衡，适于部署在智能网联汽车的环境感知系统中。

关键词: 实时目标检测, 复杂驾驶环境, DPRT-YOLO, 多尺度特征融合, Transformer

Abstract: Object detection is a fundamental task of the visual perception system in intelligent and connected vehicles (ICV), providing essential data and decision-making support for the advanced driver assistance systems (ADAS). However, in complex scenes with low illumination or severe weather, the existing vehicle-mounted object detectors fail to deliver satisfactory performance in detecting small-scale objects including the traffic signs, traffic signals and distant vehicles, leading to high false negative and false positive rates. To address this challenge, a real-time object detector (DPRT-YOLO) is developed for ICV. The proposed detector is an enhancement of the YOLOv10, making it more suitable for object detection tasks in complex driving environments. The effectiveness is verified through ablation and comparative experiments conducted on the NVIDIA embedded platform. Initially, the enhanced weighted multi-branch feature fusion network (EWMFFN) is designed to eliminate interference during feature fusion between layers, improving the ability to detect small-scale objects while maintaining a lightweight network structure. Then, the convolutional gated linear units (CGLU) and convolutional additive token mixer (CATM) are applied to construct the feature extraction module (ECAS-ViT), aiming to build long-range contextual relationships for small-scale target information while preserving lightweight characteristics. To evaluate the model’s performance in real-world computational scenarios, it is deployed on the NVIDIA Jetson Xavier NX embedded platform, with NVIDIA TensorRT FP16 quantization acceleration. Comparative experiments are conducted on the BDD100K and TT100K dataset and the results show that: (1) In terms of detection accuracy, compared with YOLOv10n and YOLO11n, the mAP@0.5 of the proposed model improves by 6.1 and 7.4 percentage points, respectively, and the mAP@0.5:0.95 improves by 3.6 and 4.2 percentage points, while the parameter count decreases by 26.1% and 34.9%, respectively. (2) In terms of inference speed, the Small and Nano versions of the proposed model achieve inference speeds of 29 FPS and 35 FPS, respectively. The experimental results demonstrate that the proposed detector outperforms reference models in complex driving environments, achieving a better balance between detection accuracy and speed, making it suitable for deployment in the environmental perception system of ICV.

Key words: real-time object detection, complex driving environments, DPRT-YOLO, multi-scale feature fusion, Transformer

董一兵, 曾辉, 李建科, 侯少杰, 石磊. DPRT-YOLO：智能网联汽车复杂驾驶环境实时目标检测器[J]. 计算机工程与应用, 2025, 61(14): 148-162.

DONG Yibing, ZENG Hui, LI Jianke, HOU Shaojie, SHI Lei. DPRT-YOLO: Real-Time Object Detector for Intelligent and Connected Vehicles in Complex Driving Environments[J]. Computer Engineering and Applications, 2025, 61(14): 148-162.

参考文献

[1] 梁华刚, 黄伟浩, 薄颖, 等. 基于多特征融合的隧道场景车辆再识别[J]. 中国公路学报, 2023, 36(8): 280-291.
LIANG H G, HUANG W H, BO Y, et al. Multifeature-fusion-based vehicle reidentification for tunnel scenes[J]. China Journal of Highway and Transport, 2023, 36(8): 280-291.
[2] 叶发茂, 张立, 袁燎, 等. DB-YOLO: 特征增强融合的双骨干YOLOv8道路缺陷检测模型[J]. 计算机工程与应用, 2024, 60(24): 260-269.
YE F M, ZHANG L, YUAN L, et al. DB-YOLO: dual backbone YOLOv8 model with feature enhancement fusion for road defect detection[J]. Computer Engineering and Applictions, 2024, 60(24): 260-269.
[3] 董一兵, 曾辉, 侯少杰. LMUAV-YOLOv8: 低空无人机视觉目标检测轻量化网络[J]. 计算机工程与应用, 2025, 61(3): 94-110.
DONG Y B, ZENG H, HOU S J. LMUAV-YOLOv8: a lightweight network for object detection in low-altitude UAV vision[J]. Computer Engineering and Applictions, 2025, 61(3): 94-110.
[4] 朱齐齐, 陈西曲. 基于改进YOLOv5的轻量级果园苹果检测算法[J]. 江苏农业科学, 2024, 52(17): 200-208.
ZHU Q Q, CHEN X Q. Lightweight orchard apple detection algorithm based on improved YOLOv5[J]. Jiangsu Agricultural Sciences, 2024, 52(17): 200-208.
[5] XU Y P, XIE Y Q, YU R, Integrated perception-communication-logistics multi-objective[J]. Journal on Communications, 2024, 45(4): 1-12.
[6] GUPTA A, MAHAUR B. An improved DV-MaxHop localization algorithm for wireless sensor networks[J]. Wireless Personal Communications, 2021, 117(3): 2341-2357.
[7] 李琳, 靳志鑫, 俞晓磊, 等. Haar小波下采样优化YOLOv9的道路车辆和行人检测[J]. 计算机工程与应用, 2024, 60(20): 207-214.
LI L, JIN Z X, YU X L, et al. Haar wavelet downsampling optimization of YOLOv9 road vehicle and pedestrian detection[J]. Computer Engineering and Applications, 2024, 60(20): 207-214.
[8] YU B Y, LI Z X, CAO Y, et al. YOLO-MPAM: efficient real-time neural networks based on multi-channel feature fusion[J]. Expert Systems with Applications, 2024, 252: 124282.
[9] 王春梅, 刘欢. YOLOv8-VSC: 一种轻量级的带钢表面缺陷检测算法[J]. 计算机科学与探索, 2024, 18(1): 151-160.
WANG C M, LIU H. YOLOv8-VSC: lightweight algorithm for strip surface defect detection[J]. Journal of Frontiers of Computer Science and Technology, 2024, 18(1): 151-160.
[10] WANG H, LIU C Y, CAI Y F, et al. YOLOv8-QSD: an improved small object detection algorithm for autonomous vehicles based on YOLOv8[J]. IEEE Transactions on Instrumentation Measurement, 2024, 73: 3379090.
[11] DING X H, ZHANG X Y, HAN J G, et al. Diverse branch block: building a convolution as an inception-like unit[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2021: 10881-10890.
[12] TAN M X, PANG R M, LE Q V. EfficientDet: scalable and efficient object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020: 10778-10787.
[13] WANG T, QU H, LIU C, et al. LLE-STD: traffic sign detection method based on low-light image enhancement and small target detection[J]. Mathematics, 2024, 12(19): 3125.
[14] LI C, GUO C, HAN L, et al. Low-light image and video enhancement using deep learning: a survey[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 44(12): 9396-9416.
[15] WANG W J, YANG W H, LIU J Y. HLA-face: joint high-low adaptation for low light face detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2021: 16190-16199.
[16] SU P, HAN H Z, LIU M, et al. MOD-YOLO: rethinking the YOLO architecture at the level of feature information and applying it to crack detection[J]. Expert Systems with Applications, 2024, 237: 121346.
[17] PAN H, GUAN S P, ZHAO X Y. LVD-YOLO: an efficient lightweight vehicle detection model for intelligent transportation systems[J]. Image and Vision Computing, 2024, 151: 105276.
[18] JIANG Z K, LIU L, ZHANG J N, et al. Dual path transformer with partition attention[J]. arXiv:2305.14768, 2023.
[19] TAN M X, LE Q V. Efficientnetv2: smaller models and faster training[C]//Proceedings of the International Conference on Machine Learning, 2021: 10096-10106.
[20] 何湘杰, 宋晓宁. YOLOv4-Tiny的改进轻量级目标检测算法[J]. 计算机科学与探索, 2024, 18(1): 138-150.
HE X J, SONG X N. Improved YOLOv4-Tiny lightweight target detection algorithm[J]. Journal of Frontiers of Computer Science and Technology, 2024, 18(1): 138-150.
[21] HUANG Y S, JIANG Z T, LAN R S, et al. Infrared image super-resolution via transfer learning and PSRGAN[J]. IEEE Signal Processing Letters, 2021, 28: 982-986.
[22] WEI H T, YU B, WANG W, et al. Adaptive enhanced detection network for low illumination object detection[J]. Mathematics, 2023, 11(10): 2404.
[23] MA X, DAI X Y, BAI Y, et al. Rewrite the stars[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2024: 5694-5703.
[24] DAUPHI Y N, FAN A, AULI M, et al. Language modeling with gated convolutional networks[C]//Proceedings of the International Conference on Machine Learning, 2017: 933-941.
[25] ZHANG T F, LI L, ZHOU Y, et al. CAS-ViT: convolutional additive self-attention vision transformers for efficient mobile applications[J]. arXiv:2408.03703, 2024.
[26] LIN T Y, DOLLAR P. GIRSHICK R, et al. Feature pyramid networks for object detection[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2017: 2117-2125.
[27] LIU S, QI L, QIN H F, et al. Path aggregation network for instance segmentation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 8759-8768.
[28] YANG Z Q, GUAN Q, ZHAO K E, et al. Multi-branch auxiliary fusion YOLO with re-parameterization heterogeneous convolutional for accurate object detection[C]//Proceedings of the Pattern Recognition and Computer Vision. Singapore: Springer Nature Singapore, 2025: 492-505.
[29] XU X Z, JIANG Y Q, CHEN W H, et al. DAMO-YOLO: a report on real-time object detection design[J]. arXiv:2211. 15444, 2022.
[30] KANG M, TING C M, TING F F, et al. ASF-YOLO: a novel YOLO model with attentional scale sequence fusion for cell instance segmentation[J]. Image and Vision Computing, 2024, 147: 105057.
[31] YANG G Y, LEI J, ZHU Z K, et al. AFPN: asymptotic feature pyramid network for object detection[C]//Proceedings of the IEEE International Conference on Systems, Man, and Cybernetics. Piscataway: IEEE, 2023: 2184-2189.
[32] HOWARD A G, ZHU M L, CHEN B, et al. MobileNets: efficient convolutional neural networks for mobile vision applications[J]. arXiv:1704.04861, 2017.
[33] HAN K, WANG Y H, TIAN Q, et al. GhostNet: more features from cheap operations[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2020: 1577-1586.
[34] YU F, CHEN H F, WANG X, et al. BDD100K: a diverse driving dataset for heterogeneous multitask learning[J]. arXiv:1805.04687, 2018.
[35] ZHU Z, LIANG D, ZHANG S H, et al. Traffic-sign detection and classification in the wild[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2016: 2110-2118.