计算机科学 ›› 2021, Vol. 48 ›› Issue (8): 106-110.doi: 10.11896/jsjkx.200700161

• 计算机图形学& 多媒体 • 上一篇    下一篇

基于关键点检测的无锚框轻量级目标检测算法

龚浩田, 张萌   

  1. 东南大学国家ASIC工程中心 南京210096
  • 收稿日期:2020-07-26 修回日期:2020-12-01 发布日期:2021-08-10
  • 通讯作者: 张萌(zmeng@seu .edu.cn)

Lightweight Anchor-free Object Detection Algorithm Based on Keypoint Detection

GONG Hao-tian, ZHANG Meng   

  1. National ASIC Engineering Center,Southeast University,Nanjing 210096,China
  • Received:2020-07-26 Revised:2020-12-01 Published:2021-08-10
  • About author:GONG Hao-tian,born in 1996,postgraduate.His main research interests include deep learning and computer vision.(220184705@seu.edu.cn)ZHANG Meng,born in 1964,Ph.D,associate professor,Ph.D supervisor.His main research interests include deep learning,machine learning,digital signal processing,digital communication systems,wireless sensor networks,digital integrated circuit design,information security and assurance,etc.

摘要: 针对基于关键点的目标检测参数量大、检测框误匹配的问题,提出一种轻量级的基于关键点检测的无锚框目标检测算法。首先将输入图片输入优化过的特征提取算法,通过级联角池化与中心池化,输出3个关键点的热力图与它们的嵌入向量;然后通过嵌入向量匹配热力图并画出检测框。文中的创新点在于将SqueezeNet中的轻量级模块firemodule适配至CenterNet,并用深度可分离卷积代替主干网的常规卷积,同时,针对CenterNet的检测框误匹配问题优化了算法输出形式与训练时的损失函数。实验结果表明,改良后的算法使得原有的CenterNet算法模型尺寸缩小为原来的1/7,同时检测精度与速度较YOLOv3,CornerNet-Lite等相同量级的算法仍有所提高。

关键词: 目标检测, 关键点, 无锚框, 轻量级, 卷积网络

Abstract: According to the large number of parameters of key-point object detection network and the problem of mismatching of bounding box,this paper proposes a lightweight key point anchor-free object detection algorithm.It inputs the image into the improved hourglass network to extract features,through the cascade corner pooling module and center pooling module,outputs three key points heatmap and their embedding vectors.At last,it matchs the key points by embedding vectors and draw the bounding box.The innovation of this paper is to applying the firemodule of SqueezeNet in the CenterNet object detection network,and replace the conventional convolution in the backbone with the depth separable convolution.At the same time,aiming at the mismatching bounding box problem in CenterNet,this algorithm adjusts the network's output and loss function.Experiment results show that the model size is reduced to 1/7 of CenterNet,while the accuracy and inference speed are still higher than the same size target detection algorithm like YOLOv3 and CornerNet-Lite.

Key words: Object detection, Key point, Anchor-free, Lightweight, Convolution network

中图分类号: 

  • TP391
[1]REDMON J,FARHADI A.Yolov3:An incremental improve-ment[J].arXiv:1804.02767.
[2]REDMON J,FARHADI A.YOLO9000:Better,Faster,Stronger[C]//IEEE Conference on Computer Vision & Pattern Recognition.2017:6517-6525.
[3]LAW H,DENG J.Cornernet:Detecting objects as paired key-points[C]//Proceedings of the European Conference on Computer Vision.2018:734-750.
[4]DUAN K W,BAI S,XIE L X.CenterNet:Keypoint Triplets for Object Detection[C]//Proceedings of the European Conference on Computer Vision.2019.
[5]REN S Q,HE K M,GIRSHICK R.Faster R-CNN:TowardsReal-Time Object Detection with Region Proposal Networks[J].IEEE Transactions on Pattern Analysis & Machine Intelligence,2017,39(6):1137-1149.
[6]GIRSHICK R.Fast R-CNN[C]//Proceedings of the 2015 IEEE International Conference on Computer Vision (ICCV).2015:1440-1448.
[7]HE K,GKIOXARI G,DOLLAR P,et al.Mask r-cnn[C]//Pro-ceedings of the IEEE International Conference on Computer Vision.2017:2961-2969.
[8]ZHOU X,WEI G,FU W L,et al.Application of deep learning in object detection[C]//2017 IEEE/ACIS 16th International Conference on Computer and Information Science (ICIS).IEEE,2017.
[9]MURPHY K P.Object detection and localization using local and global features[J].Toward Category Level Object Recognition,2006,12(1):382-400.
[10]WANG W,SHEN J,SHAO L.Video Salient Object Detectionvia Fully Convolutional Networks[J].IEEE Transactions on Image Processing,2017,27(1):38-49.
[11]YANG J,LIU Q S,ZHANG K H.Stacked Hourglass Network for Robust Facial Landmark Localisation[C]//IEEE Conference on Computer Vision & Pattern Recognition Workshops.IEEE Computer Society,2017.
[12]GIRSHICK R,DONAHUE J,DARRELL T,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2014:580-587.
[13]XIAO B,WU H,WEI Y.Simple baselines for human pose estimation and tracking[C]//ECCV.2018:472-478.
[14]SANDLER M,HOWARD A,ZHU M L,et al.MobileNetV2:Inverted Residuals and Linear Bottlenecks [C]//IEEEConfe-rence on Computer Vision and Pattern Recognition (CVPR).2018:4510-4520.
[15]IANDOLA F N,HAN S,MOSKEWICZ M W,et al.Squeeze-Net:AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size[J].arXiv,1602.07360.
[16]SZEGEDY C,LIU W,JIA Y,et al.Going Deeper with Convolutions[J].arXiv:1409.4842.
[17]HUANG G,LIU Z,LAURENS V D M,et al.Densely Connec-ted Convolutional Networks[J].arXiv:1608.06993,2016.
[18]HE K,ZHANG X,REN S,et al.Deep residual learning forimage recognition[C]//CVPR.2016.
[19]HE K,ZHANG X,REN S,et al.Identity Mappings in Deep Residual Networks[C]//European Conference on Computer Vision.Cham:Springer,2016.
[20]LAW H,TENG Y,RUSSAKOVSKY O,et al.CornerNet-Lite:Efficient Keypoint Based Object Detection[J].arXiv:1904.08900.
[1] 邢豪, 李明. 基于3D CNNS的深度伪造视频篡改检测[J]. 计算机科学, 2021, 48(7): 86-92.
[2] 宋龙泽, 万怀宇, 郭晟楠, 林友芳. 面向出租车空载时间预测的多任务时空图卷积网络[J]. 计算机科学, 2021, 48(7): 112-117.
[3] 卿来云, 张建功, 苗军. 在线异常事件检测的时序建模[J]. 计算机科学, 2021, 48(7): 206-212.
[4] 李琳, 刘学亮, 赵烨, 纪平. 结合乐高滤波器和SSD的低光照图像融合检测方法[J]. 计算机科学, 2021, 48(7): 213-218.
[5] 辛元雪, 史朋飞, 薛瑞阳. 基于区域提取与改进 LBP 特征的运动目标检测[J]. 计算机科学, 2021, 48(7): 233-237.
[6] 程思伟, 葛唯益, 王羽, 徐建. BGCN:基于BERT和图卷积网络的触发词检测[J]. 计算机科学, 2021, 48(7): 292-298.
[7] 张曼, 李杰, 朱新忠, 沈霁, 成昊天. 基于改进DCGAN算法的遥感数据集增广方法[J]. 计算机科学, 2021, 48(6A): 80-84.
[8] 宋元隆, 吕光宏, 王桂芝, 贾吾财. 基于图卷积神经网络的SDN网络流量预测[J]. 计算机科学, 2021, 48(6A): 392-397.
[9] 潘明远, 宋慧慧, 张开华, 刘青山. 学习全局引导渐进特征聚合轻量级网络的显著性目标检测[J]. 计算机科学, 2021, 48(6): 103-109.
[10] 张少钦, 杜圣东, 张晓博, 李天瑞. 融合多模态信息的社交网络谣言检测方法[J]. 计算机科学, 2021, 48(5): 117-123.
[11] 石先让, 宋廷伦, 唐得志, 戴振泳. 一种新颖的单目视觉深度学习算法:H_SFPN[J]. 计算机科学, 2021, 48(4): 130-137.
[12] 李梦荷, 许宏吉, 石磊鑫, 赵文杰, 李娟. 基于骨骼关键点检测的多人行为识别[J]. 计算机科学, 2021, 48(4): 138-143.
[13] 袁星星, 吴秦. 基于显著性特征和角度信息的遥感图像目标检测[J]. 计算机科学, 2021, 48(4): 174-179.
[14] 王丽芳, 王蕊芳, 蔺素珍, 秦品乐, 高媛, 张晋. 基于双残差超密集网络的多模态医学图像融合[J]. 计算机科学, 2021, 48(2): 160-166.
[15] 吕明琪, 洪照雄, 陈铁明. 一种融合时空关联与社会事件的交通流预测方法[J]. 计算机科学, 2021, 48(2): 264-270.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
[1] 朱淑芹,王文宏,李俊青. 针对基于感知器模型的混沌图像加密算法的选择明文攻击[J]. 计算机科学, 2018, 45(4): 178 -181 .
[2] 魏芹双,武优西,刘靖宇,朱怀忠. 基于密度约束和间隙约束的对比模式挖掘[J]. 计算机科学, 2018, 45(4): 252 -256 .
[3] 吴伟男, 刘建明. 面向低功耗无线传感器网络的动态重传算法[J]. 计算机科学, 2018, 45(6): 96 -99 .
[4] 黄铉. 特征降维技术的研究与进展[J]. 计算机科学, 2018, 45(6A): 16 -21 .
[5] 叶志斌,严波. 符号执行研究综述[J]. 计算机科学, 2018, 45(6A): 28 -35 .
[6] 姚艳玲. 2017年国际人工智能领域研究前沿的分析与研究[J]. 计算机科学, 2018, 45(9): 1 -10 .
[7] 王伟, 杨本朝, 李光松, 斯雪明. 异构冗余系统的安全性分析[J]. 计算机科学, 2018, 45(9): 183 -186 .
[8] 周明泉, 江国华. 一种基于频谱信息并结合碰集和遗传算法的缺陷定位方法[J]. 计算机科学, 2018, 45(9): 207 -212 .
[9] 陈诗军, 王慧强, 王园园, 胡海婧. 一种面向室内定位的基站选择优化方法[J]. 计算机科学, 2018, 45(10): 115 -119 .
[10] 贾伟, 华庆一, 张敏军, 陈锐, 姬翔, 王博. 基于冲突度和协同过滤的移动用户界面模式推荐[J]. 计算机科学, 2018, 45(10): 202 -206 .