一种基于视频的多目标追踪与分割算法  

A Multi-target Tracking and Segmentation Algorithm Based on Video

作　　者：黄涛 苏松源[2,3] 杜长青 诸雅琴 陈勇 HUANG Tao;SU Song-yuan;DU Chang-qing;ZHU Ya-qin;CHEN Yong(State Grid Jiangsu Electric Power Engineering Consulting Co.,Ltd.,Nanjing 210024,China;School of Cyber Science and Engineering,Southeast University,Nanjing 210096,China;School of Cyberspace Security,Southeast University,Nanjing 210000,China)

机构地区：[1]国网江苏省电力工程咨询有限公司,江苏南京210024 [2]教育部网络与信息集成重点实验室(东南大学),江苏南京210096 [3]东南大学网络空间安全学院,江苏南京210000

出　　处：《计算机技术与发展》2021年第3期95-99,共5页Computer Technology and Development

基　　金：江苏省自然基金(BK20151416);国家自然基金(61370209)。

摘　　要：基于视觉的多目标跟踪由于在智能监控、动作与行为分析、自动驾驶、虚拟现实和娱乐互动等领域都有重要的应用,近年来越来越多地成为计算机视觉领域的研究重点。并且在电力设施中对人员的活动需要实时追踪,有助于安全防护。对此,针对视频中的多目标检测与分割问题,在原有Mask-RCNN算法的基础上做了改进,引入光流分析法和视频关键帧提取技术,在不改变检测精度的同时大幅度缩短对每一帧的检测时间。实验结果表明,相较于原有Mask-RCNN算法,改进的Mask-RCNN算法大幅缩短了检测时间,对比于其他的目标追踪算法,改进的Mask-RCNN算法增强了对视频中的对象实例识别和分割的效果,分割精度有了显著提升,达到了视频里的多目标追踪的需求,并且对提高多目标场景下的目标跟踪水平具有一定的实际意义。Multi-target tracking based on vision has become an important research focus in the field of computer vision in recent years due to its important applications in the fields of intelligent monitoring,motion and behavior analysis,autonomous driving,virtual reality and entertainment interaction.And in the power facility,the activities of personnel need to be tracked in real time,which is helpful for safety protection.In this regard,the multi-target detection and segmentation problem in video is improved on the basis of the original Mask-RCNN algorithm.The optical flow analysis method and video key frame extraction technology are introduced,which greatly shortens the detection time of each frame without changing the detection accuracy.Experiment shows that compared with the original Mask-RCNN algorithm,the improved Mask-RCNN algorithm has been greatly shortened in detection time.Compared with other target tracking algorithms,the improved Mask-RCNN algorithm enhances the effect of the object instance recognition and segmentation in video,significantly improving the segmentation accuracy,which meets the needs of multi-target tracking in video,and has certain practical significance for improving the target tracking level in multi-target scenarios.

关 键 词：计算机视觉 目标追踪 光流分析法 视频关键帧 分割精度 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]