改进的YOLOv3算法在视频分析中的应用  

Application of improved YOLOv3 algorithm in video analysis

作　　者：康金龙 刘涛[2] 谢祎霖 许涛 宫胜 KANG Jinlong;LIU Tao;XIE Huilin;XU Tao;GONG Sheng(School of Economics and Management,Northwest University,Xi'an,Shaanxi 710000,China;Network and data center,Northwest University,Xi'an,Shaanxi 710000,China;Logistics group,Northwest University,Xi'an,Shaanxi 710000,China)

机构地区：[1]西北大学经济管理学院,陕西西安710000 [2]西北大学网络与数据中心,陕西西安710000 [3]西北大学后勤集团,陕西西安710000

出　　处：《信息记录材料》2022年第12期30-32,共3页Information Recording Materials

摘　　要：近年来,随着卷积神经网络的发展,目标检测的研究得到了很大的发展。然而,小的物体、紧凑和密集或高度重叠物体的识别具有挑战性。现有的方法可以很好地检测多个目标,但由于帧之间的细微变化,模型的检测效果会变得不稳定,检测结果可能会导致目标个数的下降或增加。为了解决这一问题,研究人员提出了新的YOLOv3算法YOLOv3-ANV,该算法在传统的算法上增加了判断器和优化器,来稳定检测结果序列的变化。在此基础上,采用万方体育竞赛数据集作为测试数据集,较Faster-RCNN和YOLOv3方法平均提高6.82%,具有一定的推广价值。In recent years,with the development of convolutional neural networks,object detection has been greatly developed.However,small objects,compact,dense or highly overlapping objects can be challenging to identify.The existing methods can detect multiple targets very well,but due to the subtle changes between frames,the detection effect of the model will become unstable,and the detection result may lead to the decrease or increase of the number of targets.To solve this problem,researchers proposed a new YOLOv3 algorithm,Yolov3-ANV,which added a judge and an optimizer to the traditional algorithm to stabilize the change of detection result sequence.On this basis,Wanfang square sports competition data set is used as the test data set,and the average increase is 6.82%compared with the Faster-RCNN and YOLOv3 methods,which has certain promotion value.

关 键 词：多目标行人检测 卷积神经网络 视频分析 YOLOv3 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置]