网络轻量化的实时自动驾驶目标检测算法Automatic Driving Target Detection Algorithm based on Embedded System
王荣,王锐,李俊吉,李庭鱼
摘要(Abstract)：
针对嵌入式系统中自动驾驶目标检测算法在计算量大、实时性差以及模型泛化能力不足等问题，提出了一种基于改进MobileNet-V3和Ghost模块的轻量化目标检测算法。首先，采用MobileNet-V3作为主干网络，利用深度可分离卷积技术有效减少计算量，同时保持特征提取性能。其次，在Neck部分引入Ghost模块，通过特征映射的线性转换提升多尺度信息融合能力，并显著降低模型复杂度。最后，将损失函数中的边界框回归优化为EIoULoss,以解决传统损失函数在预测框宽高变化不一致情况下的不足，从而提升模型回归性能。实验结果表明，该算法在KITTI和Gopro数据集上表现优异，平均精度(mAP)达到91.419%,召回率为97.031%,模型参数量仅为5.5 M,并在Jetson Xavier NX设备上实现了毫秒级的推理速度，充分满足嵌入式系统中实时性与检测精度的双重需求。
关键词(KeyWords)： 自动驾驶;深度学习;目标检测;轻量化;嵌入式
基金项目(Foundation): 山西省应用基础研究计划资助项目(20210302123157)
作者(Author): 王荣,王锐,李俊吉,李庭鱼
参考文献(References)：
- [1] 段续庭，周宇康，田大新，等.深度学习在自动驾驶领域应用综述 [J].无人系统技术，2021,4(6):1-27.
- [2] HONG J,HE X,DENG Z,et al.IoU-aware feature fusion R-CNN for dense object detection[J].Machine Vision and Applications,2024,35(1):3.
- [3] LYU W,LIN Q,GUO L,et al.Vehicle detection based on an imporved Faster R-CNN method[J].IEICE Transactions on Fundamentals of Electronics,Communications and Computer Sciences,2021,104(2):587-590.
- [4] DING X,LI Q,CHENG Y,et al.Local keypoint-based Faster R-CNN[J].Applied Intelligence,2020,50:3007-3022.
- [5] YANG F,HUANG L,TAN X,et al.FasterNet-SSD:a small object detection method based on SSD model[J].Signal,Image and Video Processing,2024,18(1):173-180.
- [6] YANG R,LI W,SHANG X,et al.KPE-YOLOv5:an improved small target detection algorithm based on YOLOv5[J].Electronics,2023,12(4):817.
- [7] DING P,QIAN H,BAO J,et al.L-YOLOv4:lightweight YOLOv4 based on modified RFB-s and depthwise separable convolution for multi-target detection in complex scenes[J].Journal of Real-Time Image Processing,2023,20(4):71.
- [8] XIONG S,LI B,ZHU S,et al.Spatial pyramid pooling and adaptively feature fusion based yolov3 for traffic sign detection[J].Int.Arab J.Inf.Technol.,2023,20(4):592-599.
- [9] SUN S,HAN L,WEI J,et al.ShuffleNetv2-YOLOv3:a real-time recognition method of static sign language based on a lightweight network[J].Signal,Image and Video Processing,2023,17(6):2721-2729.
- [10] LI X,SHI B,NIE T,et al.Multi-object recognition method based on improved yolov2 model[J].Information Technology and Control,2021,50(1):13-27.
- [11] 单文轩，徐永祥，白福忠，等.基于改进YOLOv3的工业指针式仪表检测 [J].工业仪表与自动化装置，2023(4):64-70.
- [12] 何凌志，王玉珏，周月娥，等.基于改进的YOLOv5算法路面检测设计[J].工业仪表与自动化装置，2023(4):93-97.
- [13] ZHOU Y,WEN S,WANG D,et al.Object detection in autonomous driving scenarios based on an improved faster-RCNN[J].Applied Sciences,2021,11(24):11630.
- [14] SONG R,AI Y,TIAN B,et al.Msfanet:A light weight object detector based on context aggregation and attention mechanism for autonomous mining truck[J].IEEE Transactions on Intelligent Vehicles,2022,8(3):2285-2295.
- [15] JIAO Z,DONG H,DIAO N.Separable CenterNet Detection Network Based on MobileNetV3—An Optimization Approach for Small-Object and Occlusion Issues[J].Mathematics,2024,12(16):2524.
- [16] LI X,WEN Z,HUA Q.Vehicle license plate recognition using Shufflenetv2 dilated convolution for intelligent transportation applications in urban Internet of Things[J].Wireless Communications and Mobile Computing,2022(1):3627246.