自动驾驶是智能交通领域研究的热点问题，也是未来交通行业的发展趋势，车前多目标检测与跟踪是自动驾驶技术的基础。交通场景中目标种类多，存在目标遮挡，影响检测与跟踪的速度和精度。因此，开展车前多目标检测与跟踪的研究，提高算法的检测精度与速度具有重要的意义。

针对YOLOv4参数量大导致车前多目标检测速度降低的问题，提出一种改进的YOLOv4目标检测算法。基于Focus操作重构输入特征层，集中宽、高信息到通道上，提高每个点的感受野；采用轻量级MobilenetV3网络替换原始YOLOv4的主干网络，采用深度可分离卷积替换特征金字塔部分的标准卷积，利用深度可分离卷积和逆残差结构简化网络模型；引入CBAM（Convolutional Block Attention Module）注意力机制，对主干网络提取到的三个特征层分别施加注意力，调整空间特征和通道特征，提高模型的检测精度和对小目标的检测效果；结合焦点损失函数和YOLOv4的损失函数，改善样本数量不均衡问题。基于Pytorch深度学习框架和公共数据集Udacity、UA-DETRAC，验证并分析改进车前多目标检测算法性能。相较于YOLOv4算法，改进YOLOv4的mAP、FPS分别提高12.85%、17 FPS，模型大小减少199 M。实验结果表明，改进算法占用内存更小、速度更快，能够准确对车前多类目标检测。

将改进的轻量型YOLOv4目标检测模型与DeepSORT目标跟踪算法相结合，采用改进YOLOv4算法对车前多目标进行检测；将卡尔曼滤波算法替换为无迹卡尔曼滤波，基于无迹卡尔曼滤波预测目标的运动状态；采用基于优先级的级联匹配算法关联当前状态与预测状态，提高目标匹配关联的准确性；针对匹配失败的目标和轨迹，采用IOU匹配再次关联。相较于原DeepSORT算法，改进算法的MOTA、MOTP、FPS分别提高4.7%、5.2%、12 FPS。实验结果表明，改进算法能够准确对车前多目标检测及跟踪，更好地适用于车前多目标检测与跟踪场景。

Automatic driving is not only a hot research topic in the field of intelligent transportation, but also the development trend of the transportation industry in the future. Multi-target detection and tracking in front of vehicles is the basis of autonomous driving technology. The speed and accuracy of detection are affected by the wide variety of targets and occlusion between targets. Therefore, it is of great significance to carry out research on multi-target detection and tracking in front of the vehicle and improve the detection accuracy and speed of the algorithm.

Aiming at the problem of low speed of multi-target detection in front of vehicle caused by large number of YOLOv4 parameters, an improved YOLOv4 target detection algorithm is proposed. The input feature layer is reconstructed based on the Focus operation, the information of width channel and height channel is concentrated to improve the receptive field of each point. The lightweight mobilenetv3 network is used to replace the backbone of the YOLOv4 and the standard convolution of the feature pyramid is replaced by the depth separable convolution. And the depth separable convolution and inverse residual structure are used to simplify the model. CBAM attention mechanism is introduced to pay attention to the three feature layers extracted from the backbone network, and adjust the features in space and channel to improve the detection accuracy of the model and the detection effect of small targets. The focal loss function is introduced to improve the loss function of YOLOv4, thereby improving the problem of unbalanced number of samples. Based on the Pytorch and public datasets Udacity and UA-DETRAC, the performance of the improved multi-target detection and tracking algorithm in front of the vehicle is verified and analyzed. Compared with YOLOv4 algorithm, mAP and FPS of improved YOLOv4 are increased by 12.85% and 17 FPS, and the model size is reduced by 199M. The experimental results show that the improved algorithm occupies less memory can detect multiple targets in front of the vehicle accurately and fastly.

The improved lightweight YOLOv4 target detection model is combined with DeepSORT target tracking algorithm, the improved YOLOv4 algorithm is used to detect multiple targets in front of the vehicle. Kalman Filtering Algorithm is replaced by Unscented Kalman Filtering to predict target motion state. Based on cascade matching algorithm, the current state and predicted state are associated to improve the accuracy of target matching correlation. For the target and trajectory that fails to match, the IOU matching is used again to correlate. Compared with DeepSORT algorithm, MOTA, MOTP and FPS are increased by 4.7%, 5.2% and 12 FPS. The experimental results show that the improved algorithm can detect and track multiple targets in front of the vehicle accurately, and can be better applied to the detection and tracking scenarios of multiple targets in front of the vehicle.

[1] 吕品,许嘉,李陶深,徐文彪.面向自动驾驶的边缘计算技术研究综述[J].通信学报,2021,42(03):190-208.

[2] 徐向阳,胡文浩,董红磊.自动驾驶汽车测试场景构建关键技术综述[J].汽车工程,2021,43(04):610-619.

[3] 钱志鸿,田春生,郭银景,王雪.智能网联交通系统的关键技术与发展[J].电子与信息学报,2020,42(01):2-19.

[4] 伍锡如,薛其威.基于激光雷达的无人驾驶系统三维车辆检测[J].光学精密工程,2022,30(04):489-497.

[5] 郑少武,李巍华,胡坚耀.基于激光点云与图像信息融合的交通环境车辆检测[J].仪器仪表学报,2019,40(12):143-151.

[6] 刘旖菲,胡学敏,陈国文,刘士豪,陈龙.视觉感知的端到端自动驾驶运动规划综述[J].中国图象图形学报,2021,26(01):49-66.

[7] 张新钰,高洪波,赵建辉.基于深度学习的自动驾驶技术综述[J].清华大学学报(自然科学版),2018,058(004):438-444.

[8] 董文轩,梁宏涛,刘国柱,胡强,于旭.深度卷积应用于目标检测算法综述[J].计算机科学与探索,2022,16(05):1025-1042.

[9] Uijlings J.,Sande K.,Gevers T.,et al. Selective Search for Object Recognition[J]. International Journal of Computer Vision, 2013,104(2):154-171.

[10] Ming-Ming, Cheng, Yun, et al. BING: Binarized normed gradients for objectness estimation at 300fps[J]. Computational Visual Media, 2019, 5(01):4-21.

[11] Moranduzzo T. ,Melgani F. , Automatic Car Counting Method for Unmanned Aerial Vehicle Images[J]. IEEE Transactions on Geoscience & Remote Sensing, 2013, 52(3):1635-1647.

[12] Cheng H. , Weng C., Chen Y. Vehicle Detection in Aerial Surveillance Using Dynamic Bayesian Networks[J]. 2012, 21(4):2152-2159.

[13] Mattyus, Gellert, Liu, et al. Fast Multiclass Vehicle Detection on Aerial Images[J]. IEEE geoscience and remote sensing letters, 2015, 12(9):1938-1942.

[14] Alom M. , Taha T., Yakopcic C. , et al. The History Began from AlexNet: A Comprehensive Survey on Deep Learning Approaches[J].2018, 12(3): 1803-1842.

[15] Girshick R., Donahue J., Darrell T., et al. Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation[J]. IEEE Computer Society, 2014,24(3): 580-587.

[16] Kaiming He, Xiangyu,et al.Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition[J].IEEE Transactions on Pattern Analysis & Machine Intelligence, 2015, 37(9): 1904-1916.

[17] 曹诗雨, 刘跃虎, 李辛昭. 基于Fast R-CNN的车辆目标检测[J]. 中国图象图形学报, 2017, 22(5):7-19.

[18] Ren S., He K., Girshick R. , et al. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 39(6):1137-1149.

[19] Wang Q. , Zhou L. , Yao Y. , et al. An Interconnected Feature Pyramid Networks for object detection[J].Journal of Visual Communication and Image Representation, 2021,79(3):1032-1047.

[20] Fan Z. , Xia W. , Liu X. , et al. Detection and segmentation of underwater objects from forward-looking sonar based on a modified Mask RCNN[J]. Signal Image and Video Processing, 2021,15(6):1135-1143.

[21] 黄健,张钢. 深度卷积神经网络的目标检测算法综述[J]. 计算机工程与应用, 2020, 56(17):12-23.

[22] 魏湧明, 全吉成, 侯宇青阳. 基于YOLO v2的无人机航拍图像定位研究[J]. 激光与光电子学进展, 2017, 54(11):10-19.

[23] Mazzara M. . Detection of Small Size Traffic Signs Using Regressive Anchor Box Selection and DBL Layer Tweaking in YOLOv3[J]. Applied Sciences, 2021, 11(23): 354-361.

[24] Fan S. , Liang X. , Huang W. , et al. Real-time defects detection for apple sorting using NIR cameras with pruning-based YOLOV4 network[J]. Computers and Electronics in Agriculture, 2022, 193(3):106-124.

[25] Popescu D. E . An Integrated Approach for Monitoring Social Distancing and Face Mask Detection Using Stacked ResNet-50 and YOLOv5[J]. Electronics, 2021, 10(23): 156-172.

[26] Xu F. , Li X. , Yang H. , et al. TE-YOLOF: Tiny and efficient YOLOF for blood cell detection[J]. Biomedical Signal Processing and Control, 2022, 73(11):103-116.

[27] 谢俊章,彭辉, 唐健峰.改进YOLOv4的密集遥感目标检测[J].计算机工程与应用,2021,57(22):247-256.

[28] Cai Y. , Luan T. , Gao H. , et al. YOLOv4-5D: An Effective and Efficient Object Detector for Autonomous Driving[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 7(19):163-185.

[29] Chen W. , Qiao Y. , Li Y . Inception-SSD: An improved single shot detector for vehicle detection[J]. Journal of Ambient Intelligence and Humanized Computing, 2020, 27(5) :46-61.

[30] Yang Y U, Wang Q, University S L. Research on Obstacle Detection of Unmanned Vehicle Based on DSSD[J]. Journal of Shenyang Ligong University, 2018,37(4):77-89.

[31] Liang X. , Zhang J. , Zhuo L., et al. Small Object Detection in Unmanned Aerial Vehicle Images Using Feature Fusion and Scaling-Based Single Shot Detector with Spatial Context Analysis[J].IEEE Transactions on Circuits and Systems for Video Technology, 2019,4(12) :1758-1770.

[32] Lin T Y , Goyal P. , Girshick R. , et al. Focal Loss for Dense Object Detection[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 42(2):318-327.

[33] 龚红,杨发顺,丁召.基于模糊背景加权的Mean Shift目标跟踪算法[J].电子科技大学学报, 2019, 48(3):402-408.

[34] 刘洪彬,常发亮.权重系数自适应光流法运动目标检测[J].光学精密工程,2016, 24(2):9-21.

[35] 杨鹏生,吴晓军,张玉梅.改进扩展卡尔曼滤波算法的目标跟踪算法[J].计算机工程与应用,2016, 52(5):5-23.

[36] 王法胜,鲁明羽,赵清杰,等.粒子滤波算法[J].计算机学报, 2014,37(8):16-31.

[37] 王霞,王光磊,李艳,等.改进的基于特征融合MOSSE冠脉目标追踪算法[J].电子测量与仪器学报, 2021, 35(9):7-22.

[38] Lan Y. , Ma P. , Xu A. , et al. A Visual Tracking Method Based on an Adaptive Overlapping Correlation Filter for Robotic Real-Time Cognitive Imaging[J]. Wireless Communications and Mobile Computing, 2020,3(16):1-8.

[39] Henriques J F , Caseiro R. , Martins P. , et al. High-Speed Tracking with Kernelized Correlation Filters[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2015, 37(3):583-596.

[40] Danelljan M. , Khan F S , Felsberg M, et al. Adaptive Color Attributes for Real-Time Visual Tracking[C]// IEEE Conference on Computer Vision & Pattern Recognition. IEEE, 2014：589-201.

[41] J. Shao, Du B. , Wu C. , et al. Can We Track Targets From Space? A Hybrid Kernel Correlation Filter Tracker for Satellite Video[J]. IEEE Transactions on Geoscience and Remote Sensing, 2019, 23(99):1-13.

[42] 马珺,王昱皓.结合自适应更新策略和再检测技术的跟踪算法[J].计算机工程与应用, 2021, 57(9):8-17.

[43] Bao F., Cao Y., Zhang S., Lin B.Using segmentation with multi-scale selective keneral for visual object tracking[J].IEEE Signal Processing Letters,2011 42(10):132-144.

[44] Martin, Danelljan, Gustav, et al. Discriminative Scale Space Tracking[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(8):1561-1575.

[45] Fan R. , Fan J. , Liu Q. Real-time High-confidence Update Complementary Learner Tracking[J]. Computer Science, 2019,31(6):145-161.

[46] Wibowo S A , Lee H., Kim E K , et al. Convolutional Shallow Features for Performance Improvement of Histogram of Oriented Gradients in Visual Object Tracking[J]. Mathematical Problems in Engineering, 2017, 2017(7):1-9.

[47] Danelljan M. , Bhat G. , Khan F. S. , et al. ECO: Efficient Convolution Operators for Tracking[J]. IEEE Computer Society, 2016,28(10):201-223.

[48] Li S. , Ma H. . A Siamese inception architecture network for person re-identification[J]. Machine Vision & Applications, 2017, 28(7):725-736.

[49] Kuai Y. , Wen G. , Li D. Masked and dynamic Siamese network for robust visual tracking[J]. Information Sciences, 2019, 50(3):169-182.

[50] Huang Z. , Zhao H. , Zhan J. , et al. A multivariate intersection over union of SiamRPN network for visual tracking[J]. The Visual Computer, 2021,24(8):1-12.

[51] 王玲,王家沛,王鹏,等.融合注意力机制的孪生网络目标跟踪算法研究[J].计算机工程与应用, 2021, 57(8):6-20.

[52] 李勇,杨德东,韩亚君,等.融合扰动感知模型的孪生神经网络目标跟踪[J].光学学报, 2020, 40(4):120-131.

[53] Jin L. S. , Hua Q. , B. C. Guo, et al. Multi-target tracking of vehicles based on optimized DeepSort[J]. Journal of Zhejiang University SCIENCE, 2021,55(6):1056-1064.

[54] Jain M., Subramanyam A. V., Denman S. , et al. LSTM guided ensemble correlation filter tracking with appearance model pool[J]. Computer Vision and Image Understanding, 2020, 195(8):102-135.

[55] Technicolor T., Related S., Technicolor T. , et al. ImageNet Classification with Deep Convolutional Neural Networks [J].Communications of the ACM,2017,60(6):84-90.

[56] Cha Y. J. , Choi W. Deep Learning-Based Crack Damage Detection Using Convolutional Neural Networks[J]. Computer-Aided Civil and Infrastructure Engineering, 2017,32(5):361-378.

[57] Srivastava N., Hinton G., Krizhevsky A., et al. Dropout: A Simple Way to Prevent Neural Networks from Overfitting[J]. Journal of Machine Learning Research, 2014, 15(1):1929-1958.

[58] 王立辉, 杨贤昭, 刘惠康,等.基于GhostNet与注意力机制的行人检测跟踪算法[J].数据采集与处理, 2022, 37(1):14-27.

[59] Yang Z. . Activation Function: Cell Recognition Based on YoLov5s/m[J]. Journal of Computer and Communications, 2021, 9(12):16-34.

[60] Kaiming, He, Xiangyu, et al. Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition[J].IEEE Transactions on Pattern Analysis & Machine Intelligence, 2015,37(9):1904-1916.

[61] Wang Q. , Zhou L., Yao Y., et al. An Interconnected Feature Pyramid Networks for object detection[J]. Journal of Visual Communication and Image Representation, 2021,56(3):103-122

[62] Li F., Li W., Qin S., et al. MDFA-Net: Multiscale dual-path feature aggregation network for cardiac segmentation on Multi-sequence Cardiac MR[J]. Knowledge-Based Systems, 2021, 215(10):167-181.

[63] Kcab C. , Xmab C. , Wei W. , et al. A modified YOLOv3 model for fish detection based on MobileNetv1 as backbone-ScienceDirect[J]. Aquacultural Engineering, 2020,37(10):154-163.

[64] Rajendran P. K. , Mishra S. , Vecchietti L. F., et al. RELMOBNET: A Robust Two-Stage End-To-End Training Approach For MobilenetV3 Based Relative Camera Pose Estimation[J]. 2022,23(7):87-104.

[65] Men H. ,Yuan H. ,Shi Y., et al. A Residual Network with Attention Module for Hyperspectral Information of Recognition to Trace the Origin of Rice[J]. Spectrochimica Acta Part A Molecular and Biomolecular Spectroscopy, 2021,23(8):107-125.

[66] Lin T., Goyal P., Girshick R , et al. Focal Loss for Dense Object Detection[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2017, 28(7):199-213.

[67] Jia Ruiyu,Li Yugong.K-means algorithm of clustering number and centers self-determination[J].Computer Engineering and Applications,2018,54(7):152-158.

[68] Wen L., Du D. , Cai Z., et al. UA-DETRAC: A New Benchmark and Protocol for Multi-Object Detection and Tracking[J]. Computer Science, 2015,32(10):146-171.

[69] 金立生,华强,郭柏苍,谢宪毅,闫福刚,武波涛.基于优化DeepSort的前方车辆多目标跟踪[J].浙江大学学报(工学版),2021,55(06):1056-1064.

[70] 鲍水达,张安,毕文豪.快速强跟踪UKF算法及其在机动目标跟踪中的应用[J].系统工程与电子技术,2018,40(06):1189-1196.

[71] 徐涛,马克,刘才华.基于深度学习的行人多目标跟踪方法[J].吉林大学学报(工学版),2021,51(01):27-38.