吉林大学学报(工学版) ›› 2024, Vol. 54 ›› Issue (10): 2799-2806.doi: 10.13229/j.cnki.jdxbgxb.20230474

• 交通运输工程·土木工程 • 上一篇    

融合GPNet与图像多尺度特性的红外-可见光道路目标检测优化方法

孙文财1(),胡旭歌1,杨志发1,2,3(),孟繁雨2,孙微3   

  1. 1.吉林大学 交通学院,长春 130022
    2.中国第一汽车集团有限公司 产品策划及项目管理部,长春 130013
    3.长春汽车工业高等专科学校,长春 130031
  • 收稿日期:2023-05-12 出版日期:2024-10-01 发布日期:2024-11-22
  • 通讯作者: 杨志发 E-mail:swcai@163.com;yangzf@jlu.edu.cn
  • 作者简介:孙文财(1981-),男,教授,博士.研究方向:车辆运行监控预警,智能网联与自动驾驶.E-mail: swcai@163.com
  • 基金资助:
    国家自然科学基金项目(51978310);吉林省交通运输创新发展支撑项目(2020-1-12)

Optimization of infrared-visible road target detection by fusing GPNet and image multiscale features

Wen-cai SUN1(),Xu-ge HU1,Zhi-fa YANG1,2,3(),Fan-yu MENG2,Wei SUN3   

  1. 1.College of Transportation,Jilin University,Changchun 130022,China
    2.Product Planning and Project Management Department,China FAW Group Co. ,Ltd. ,Changchun 130013,China
    3.Changchun Automobile Industry Institute,Changchun 130031,China
  • Received:2023-05-12 Online:2024-10-01 Published:2024-11-22
  • Contact: Zhi-fa YANG E-mail:swcai@163.com;yangzf@jlu.edu.cn

摘要:

为提高道路交通安全领域中的道路目标检测的精度,本文借鉴图像融合技术中多尺度特征图像融合思想,融合和GPNet中Ghost瓶颈模块,实现了融合质量和较小算法复杂度的平衡,建立了一种红外和可见光融合及目标检测网络。网络分为选择性图像融合模块、轻量化目标检测模块和融合质量及检测精度判别网络3个部分。在白天、夜间和特殊天气(雨、雾等)下,平均车速30~40 km/h的城市工况下进行3组试验作为数据集,实验结果表明:平均梯度最高提升5.648 81、交叉熵提升了0.936 68、边缘强度提升了56.945 7、信息熵提升了0.925 208 781、互信息提升了1.000 548 571、峰值信噪比提升了3.053 893 252、Qab提升了0.342 882 208、Qcb提升了0.208 983 81以及均方误差降低0.08。轻量化目标检测网络输出的AP、mAP和Recall均为最优水平,验证了红外和可见光技术应用在道路障碍物检测方面的优势。

关键词: 交通运输系统工程, 计算机视觉, 红外和可见光图像融合, 多尺度图像融合, YOLOv5目标检测

Abstract:

In order to improve the accuracy of road target detection in the field of road traffic safety, an innovative infrared and visible fusion and detection network is established by borrowing the idea of multi-level feature image fusion for fusion in image fusion technology and the idea of Ghost bottleneck module building in GPNet to reduce the complexity of the algorithm. The network is divided into three parts: selective image fusion module, lightweight target detection module and fusion quality and detection accuracy discriminative network. Three sets of experiments were conducted as data sets under urban working conditions with an average vehicle speed of 30-40 km/h in daytime, nighttime and special weather (rain, fog, etc.). Experimental results: the highest average gradient lift of 5.648 81, cross-entropy of 0.936 68, edge strength of 56.945 7, information entropy of 0.925 208 781, mutual information of 1.000 548 571, peak signal-to-noise ratio 3.053 893 252, Qab0.342 882 208, Qcb0.208 983 81 and mean square error reduction of 0.08. The AP, mAP and Recall of the output of the lightweight target detection network are all at the optimal level, which verifies the advantages of the innovative application of infrared and visible light technologies for road obstacle detection.

Key words: engineering of communication and transportation, computer vision, infrared and visible image fusion, YOLOv5 target detection

中图分类号: 

  • U492.8

图1

网络架构"

图2

选择性图像融合模块"

图3

特征提取网络结构"

图4

实验路线"

图5

红外摄像头和可见光摄像头对比"

图6

直方图均衡化后RGB图像与原图像对比及分布直方图"

图7

暗通道去雾前及去雾后对比"

表1

选择性图像预处理实验结果"

方 法图像总数/张选择频次PSNR均值SSIM均值MSE均值综合指标均值
直方图均衡化350177390.898 60.03539.863 6
暗通道先验350173410.879 30.04441.835 3

图8

典型图像融合策略实验图像对比"

图9

本文融合框架融合实验"

图10

融合方式效果对比"

表2

各种融合图像策略平均用时对比"

方法用时/(s·帧-1方法用时/(s·帧-1
ADF4.56FPDE9.41
CBF68.11GFCE10.56
CNN140.51GTF16.16
HMSD_GF7.96LatLRR1 036.78
Hybrid_MSD36.46MGFF3.64
IFEVIP1.48MSVD6.78
本文2.65

表3

YOLOv5检测效果"

方法

融合

车AP/%人AP/%平均AP/%Recall/%mAP50/%
IR75.484.379.8544.3249.7
VI70.180.775.4045.3146.5
CNN77.288.382.7546.2360.7
ADF77.582.379.9047.2151.7
LatLRR77.378.477.8547.8960.9
Hybrid_MSD67.378.773.0045.9861.7
IFEVIP77.179.178.145.2157.6
MGFF67.582.675.0544.9958.7
MSVD67.481.074.2047.786
TIF67.784.576.146.5264.3
本文83.288.986.0549.3465.7
1 纪嘉树.基于多传感器融合的无人驾驶环境感知技术研究[D].济南: 山东大学能源与动力工程学院,2022.
Ji Jia-Shu. Research on environment perception technology of unmanned driving based on multi-sensor fusion[D]. Ji'nan: School of Energy and Power Engineering,Shandong University, 2022.
2 张博雅. 红外与可见光数字图像融合技术研究[D].南京: 南京理工大学电子工程与光电技术学院, 2015.
Zhang Bo-Ya. Infrared and visible light digital image fusion technology research[D]. Nanjing: School of Electronic Engineering and Optoelectronic Technology, Nanjing University of Science & Technology, 2015.
3 刘畅. 基于深度卷积网络的多模图像融合与质量提升方法研究[D].衡阳: 南华大学电气工程学院, 2022.
Liu Chang. Research on multimodal image fusion and quality improvement method based on deep convolutional network[D]. Hengyang: School of Electrical Engineering,University of South China, 2022.
4 张婉婉. 基于深度学习的遥感图像融合算法研 究[D]. 烟台: 山东工商学院计算机科学与技术学院, 2022.
Zhang Wan-wan. Research on remote sensing image fusion algorithm on deep learning[D]. Yantai: College of Computer Science and Technology,Shandong Technology and Business University, 2022.
5 Fang L, Wang X, Wan Y. Adaptable active contour model with applications to infrared ship target segmentation[J].Journal of Electronic Imaging,2016,25(4):1093-1100.
6 Bavirisetti D P, Dhuli R, Fusion of infrared and visible sensor images based on anisotropic diffusion and karhunen-loeve transform[J]. IEEE Sensors Journal, 2016, 16(1): 203-209.
7 Kumar B K S. Image fusion based on pixel significance using cross bilateral filter[J]. Signal, Image and Video Processing, 2013, 9(5): 1193-1204.
8 Liu Y, Chen X, Cheng J, et al. Infrared and visible image fusion with convolutional neural networks[J]. International Journal of Wavelets, Multiresolution and Information Processing, 2018, 16(3): No.1850018.
9 高雪琴, 刘刚, 肖刚, 等. 基于FPDE的红外和可见光图像融合算法[J]. 自动化学报, 2020, 46(4): 796-804.
Gao Xue-Qin, Liu Gang, Xiao Gang, et al. Fusion algorithm of infrared and visible images based on FPDE[J]. Acta Automatica Sinica, 2020, 46(4): 796-804.
10 Zhou Z, Dong M, Xie X, et al. Fusion of infrared and visible images for night-vision context enhancement[J]. Appl Opt, 2016, 55(23): 6480-6490.
11 Zhang Y, Zhang L, Bai X Z, et al. Infrared and visual image fusion through infrared feature extraction and visual information preservation[J]. Infrared Physics & Technology,2017. 83: 227-237.
12 Bavirisetti D P, Xiao G, Zhao J, et al. Multi-scale guided image and video fusion[J]. Circuits, Systems, and Signal Processing, 2019, 38: 5576-5605.
13 Vasiliki B, Loukas B, Stefanos K, et al. Generating graph-inspired descriptors by merging ground-level and satellite data for robot localization[J]. Cybernetics and Systems, 2022, 13(1): 697-715.
14 Bavirisetti D P, Dhuli R. Two-scale image fusion of visible and infrared images using saliency detection[J]. Infrared Physics & Technology, 2016, 76: 52-64.
15 Redmon J, Divvala S, Girshick R, et al. You only look once: unified, real-time object detection[J/OL].[2015-01-10]. .
16 高明华,杨璨.基于改进卷积神经网络的交通目标检测方法[J].吉林大学学报: 工学版,2022,52(6):1353-1361.
Gao Ming-hua, Yang Can.Traffic target detection method based on improved convolution neural network[J].Journal of Jilin University(Engineering and Technology Edition), 2022, 52(6): 1353-1361.
[1] 张娜,陈峰,王剑坡,朱亚迪. 基于时空序列相似性的城轨乘客出行模式识别[J]. 吉林大学学报(工学版), 2024, 54(9): 2588-2599.
[2] 周锡浈,宫贺,李敦敦,季彦婕,严杰. 建成环境对路内停车泊位使用率的非线性影响模型[J]. 吉林大学学报(工学版), 2024, 54(9): 2520-2530.
[3] 严利鑫,曾涛,贺宜,郭军华,胡鑫辉. 共驾型智能车辆人机接管行为序列编码与解析[J]. 吉林大学学报(工学版), 2024, 54(9): 2547-2556.
[4] 曲昭伟,李霖,陈永恒,吴场建. 长区间掉头车辆特性分析及其安全评价[J]. 吉林大学学报(工学版), 2024, 54(8): 2206-2213.
[5] 才华,寇婷婷,杨依宁,马智勇,王伟刚,孙俊喜. 基于轨迹优化的三维车辆多目标跟踪[J]. 吉林大学学报(工学版), 2024, 54(8): 2338-2347.
[6] 朱圣杰,王宣,徐芳,彭佳琦,王远超. 机载广域遥感图像的尺度归一化目标检测方法[J]. 吉林大学学报(工学版), 2024, 54(8): 2329-2337.
[7] 程国柱,盛林,王浩宇,冯天军. 考虑右转车二次冲突的信号交叉口行人过街安全评价方法[J]. 吉林大学学报(工学版), 2024, 54(7): 1903-1912.
[8] 何永明,权聪,魏堃,冯佳,万亚楠,陈世升. 超高速公路车路协同路侧单元感知融合方法[J]. 吉林大学学报(工学版), 2024, 54(7): 1923-1934.
[9] 孙铭会,薛浩,金玉波,曲卫东,秦贵和. 联合时空注意力的视频显著性预测[J]. 吉林大学学报(工学版), 2024, 54(6): 1767-1776.
[10] 王宇,赵凯. 基于亚像素定位的人体姿态热图后处理[J]. 吉林大学学报(工学版), 2024, 54(5): 1385-1392.
[11] 王殿伟,张池,房杰,许志杰. 基于高分辨率孪生网络的无人机目标跟踪算法[J]. 吉林大学学报(工学版), 2024, 54(5): 1426-1434.
[12] 张明业,杨敏,黎彧,黄世玉,李清韵. 考虑有序充电策略的多车型电动公交调度优化[J]. 吉林大学学报(工学版), 2024, 54(5): 1293-1301.
[13] 秦雅琴,钱正富,谢济铭. 协同换道避障模型和轨迹数据驱动的车辆协同避障策略[J]. 吉林大学学报(工学版), 2024, 54(5): 1311-1322.
[14] 高云龙,任明,吴川,高文. 基于注意力机制改进的无锚框舰船检测模型[J]. 吉林大学学报(工学版), 2024, 54(5): 1407-1416.
[15] 涂辉招,王万锦,乔鹏,郭静秋,鹿畅,吴海飞. 自动驾驶卡车路测安全员接管干预行为解析[J]. 吉林大学学报(工学版), 2024, 54(3): 727-740.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!