RoadNetv2：高速公路弱小异物目标实时检测算法

doi:10.3778/j.issn.1002-8331.2207-0166

摘要/Abstract

摘要： 高速异物易引起汽车失控，造成交通安全事故。针对现阶段高速异物数据集复杂度低以及高速异物检测算法存在检测精度低、浮点计算量高等问题，提出了模拟场景数据集扩充方法和一种名为RoadNetV2的高速异物检测算法。模拟场景数据集扩充方法利用相似数据集进行场景模拟扩充。roadnetv2高速异物检测算法采用了light-focus浅层信息增强模块和C3_CD特征提取模型作为backbone主要组件，采用CoordConv与自定义Conv组合的方法降低neck部分的复杂度，采用多权值平衡计算策略辅助Alpha-CIOU进行弱小目标高效位置回归。实验证明RoadNetV2高速异物检测算法相比于现阶段同系列YOLO算法，浮点计算量最多降低了14.54×109仅为12.4×109，mAP最高提升了3.5个百分点达到61.1%，权重文件仅为8.70 MB减少了4.98 MB。RoadNetV2高速异物检测算法满足嵌入式边缘设备以及移动端设备的部署要求，搭配自主研发的巡检设备，可以一定程度上解决人工巡检所产生工作懈怠的情况。

关键词: 高速异物, 弱小目标, 轻量化神经网络, 复杂背景, 边缘计算

Abstract: Highway abandoned objects easily cause car out of control, resulting in traffic accidents. In order to solve the problems of low complexity of highway abandoned objects dataset and low detection accuracy and high FLOPS calculation of highway abandoned objects detection algorithm at present, a simulation scenario dataset expansion method and a RoadNetV2 highway abandoned objects detection algorithm are proposed. Simulation scenario dataset expansion method uses similar datasets to expand the scene simulation. RoadNetV2 highway abandoned objects detection algorithm adopts light-focus shallow information enhancement module and C3_CD feature extraction model as main components of Backbone, adopts CoordConv and custom Conv combination method to reduce the complexity of neck, and adopts multi-weight balance calculation strategy to assist Alpha-CIOU to weak objects efficient position regression. Experimental results show that compared with the current series of YOLO algorithms, the FLOPS calculation of RoadNetV2 highway abandoned objects detection algorithm is reduced by 14.54×109 at most to 12.4×109, and mAP（mean average precision） is improved by 3.5?percentage points to 61.1%, weight files are only 8.70 MB less by 4.98 MB. RoadNetV2 highway abandoned objects detection algorithm can meet the deployment requirements of embedded edge devices and mobile devices. Combined with self-developed inspection equipment, RoadNetV2 can solve the slack caused by manual inspection to a certain extent.

Key words: highway abandoned objects, weak objects, lightweight neural network, complex background, edge computing

朱晓峰, 李林, 张德津, 罗文婷. RoadNetv2：高速公路弱小异物目标实时检测算法[J]. 计算机工程与应用, 2023, 59(1): 317-324.

ZHU Xiaofeng, LI Lin, ZHANG Dejin, LUO Wenting. RoadNetv2：Real-Time Algorithm for Highway Weak Abandoned Objects Detection[J]. Computer Engineering and Applications, 2023, 59(1): 317-324.

参考文献

[1] 吴宏涛，刘力源，孟颖，等.动态多特征融合的道路遗洒物威胁度分析方法[J].计算机科学，2020，47（S1）：196-205.
WU H T，LIU L Y，MENG Y.et al.Novel threat degree analysis method for scattered objects in road traffic based on dynamic multi-feature fusion[J].Computer Science，2020，47（S1）：196-205.
[2] 汪贵平，马力旺，郭璐，等.高速公路抛洒物事件图像检测算法[J].长安大学学报（自然科学版），2017，37（5）：8188.
WANG G P，MA L W，GUO L.et al.Image detection algorithm for incident of discarded things in highway[J].Journal of Chang’an University（Natural Science Edition），2017，37（5）：8188.
[3] 李清瑶，邹皓，赵群，等.基于帧间差分自适应法的车辆抛洒物检测[J].长春理工大学学报（自然科学版），2018，41（4）：108-113.
LI Q Y，ZOU H，ZHAO Q.et al.Vehicles throwing detection based on inter-frame difference adaptive method[J].Journal of Changchun University of Science and Technology（Natural Science Edition），2018，41（4）：108-113.
[4] 夏莹杰，欧阳聪宇.面向高速公路抛洒物检测的动态背景建模方法[J].浙江大学学报（工学版），2020，54（7）：1249-1255.
XIA Y J，OUYANG C Y.Dynamic image background modeling method for detecting abandoned objects in highway[J].Journal of Zhejiang University（Engineering Science），2020，54（7）：1249-1255.
[5] 金瑶，张锐，尹东.城市道路视频中小像素目标检测[J].光电工程，2019，46（9）：76-83.
JIN Y，ZHANG R，YIN D.Object detection for small pixel in urban roads videos[J].Opto-Electronic Engineering，2019.46（9）：76-83.
[6] 章悦，张亮，谢非，等.基于实例分割模型优化的道路抛洒物检测算法[J].计算机应用，2021，41（11）：3228-3233.
ZHANG Y，ZHANG L，XIE F，et al.Road abandoned object detection algorithm based onoptimized instance segmentation model[J].Journal of Computer Applications，2021，41（11）：3228-3233.
[7] SANDLER M，HOWARD A，ZHU M，et al.Mobilenetv2：Inverted residuals and linear bottlenecks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2018：4510-4520.
[8] MA N，ZHANG X，ZHENG H T，et al.ShuffleNet V2：Practical guidelines for efficient CNN architecture design[C]//Proceedings of the European Conference on Computer Vision，2018：116-131.
[9] 郭宇阳，胡伟超，戴帅，等.面向路侧交通监控场景的轻量车辆检测模型[J].计算机工程与应用，2022，58（6）：192-199.
GUO Y Y，HU W C，DAI S，et al.Lightweight vehicle detection model for roadside traffic monitoring scenarios[J].Computer Engineering and Applications，2022，58（6）：192-199.
[10] 丁培，阿里甫·库尔班，耿丽婷，等.自然环境下实时人脸口罩检测与规范佩戴识别[J].计算机工程与应用，2021，57（24）：268-275.
DING P，ALIFU·KUERBAN，GENG L T，et al.Real-time face mask detection and standard wearing recognition method in natural environment[J].Computer Engineering and Applications，2021，57（24）：268-275.
[11] LI Y H，YAO T，PAN Y W，et al.Contextual transformer networks for visual recognition[J].arXiv：2107.12292，2021.
[12] LIU S，QI L，QIN H，et al.Path aggregation network for instance segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition，2018：8759-8768.
[13] LIU R N，LEHMAN J，MOLINO P，et al.An intriguing failing of convolutional neural networks and the CoordConv solution[C]//Advances in Neural Information Processing Systems，2018：9628-9639.
[14] HE J，ERFANI S，MA X，et al.Alpha-IOU：A family of power intersection over union losses for bounding box regression[C]//Advances in Neural Information Processing Systems，2021：20230-20242.
[15] HAN K，WANG Y，TIAN Q，et al.GhostNet：More features from cheap operations[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition，2020：1580-1589.