第14卷第2期 智能系统学报 Vol.14 No.2 2019年3月 CAAI Transactions on Intelligent Systems Mar.2019 D0:10.11992/tis.201709003 网络出版地址：http:/kns.cnki.net/kcms/detail/23.1538.tp.20180417.1832.013.html 一种基于Multi-Egocentric视频运动轨迹重建的 多目标跟踪算法 欧伟奇2，尹辉2，许宏丽2，刘志浩2 (1.北京交通大学计算机与信息技术学院，北京100044,2.北京交通大学交通数据分析与挖掘北京市重点实 验室，北京100044) 摘要：Egocentric视频具有目标运动剧烈、遮挡频繁、目标尺度差异明显及视角时变性强的特点，给目标跟踪 任务造成了极大的困难。本文从重建不同视角Egocentric视频中各目标的运动轨迹出发，提出一种基于Multi-- Egocentric视频运动轨迹重建的多目标跟踪算法，该方法基于多视角同步帧之间的单应性约束解决目标遮挡和 丢失问题，然后根据多视角目标空间位置约束关系通过轨迹重建进一步优化目标定位，并采用卡尔曼滤波构建 目标运动模型优化目标运动轨迹，在BJMOT、EPLF-campus4数据集上的对比实验验证了本文算法在解决 Multi--Egocentric视频多目标跟踪轨迹不连续问题的有效性。 关键词：Multi-egocentric视频；轨迹重建；多目标跟踪；单应性约束；对极几何约束；空间重构；卡尔曼滤波：运动 模型 中图分类号：TP391.4文献标志码：A文章编号：1673-4785(2019)02-0246-08 中文引用格式：欧伟奇，尹辉，许宏丽，等.一种基于Multi-Egocentric视频运动轨迹重建的多目标跟踪算法J.智能系统学报， 2019,14(2):246-253. 英文引用格式：OU Weiqi,.YIN Hui,XU Hongli,etal.A multi-object tracking algorithm based on trajectory reconstruction on multi-egocentric videoJl.CAAI transactions on intelligent systems,2019,14(2):246-253. A multi-object tracking algorithm based on trajectory reconstruction on multi-egocentric video OU Weiqi2,YIN Hui2,XU Hongli2,LIU Zhihao2 (1.Department of Computer and Information Technology,Beijing Jiaotong University,Beijing 100044,China;2.Beijing Key Lab of Transportation Data Analysis and Mining,Beijing Jiaotong University,Beijing 100044,China) Abstract:In egocentric video,objects have the characteristics of violent motion,frequent occlusion,so it brings much trouble to carrying out the tracking task.In this paper,we propose a multi-object tracking algorithm based on the mo- tion trajectory reconstruction of multi-egocentric video from different visual angles egocentric videos.First,this method is based on the homographic constraint of multi-view synch frames to fix position of occluded and missing object. Second,using the relative position constraint relation of multi-angle target,the trajectory is reconstructed to locate the target position.Meanwhile,the trajectory of the object is optimized by constructing the motion model of object.Then the continuous trajectory of the object is obtained and the problem of the discontinuity trajectory in multi-object track- ing is resolved.In the end,the performance of proposed method has been verified by using the multi-view video dataset which is created by us. Keywords:Multi-Egocentric video;trajectory reconstruction;multi-object tracking;homographic constraint;epipolar geometry constraint;spatial reconstruction;Kalman filter;motion model 收稿日期：2017-09-05.网络出版日期：2018-04-18. 目标跟踪是计算机视觉重要研究领域之一， 基金项目：国家自然科学基金项目(61472029,61473031)：科技 部国家重点研发计划项目(2017NFB1201104,2016YFB 在智能交通、运动分析、行为识别、人机交互四等 I200100):中央高校基本科研业务费专项资金项目 (2016JBZ005). 方面具有广泛应用。随着可穿戴式相机的普及， 通信作者：尹辉.E-mail:hyin@bjtu.edu.cn. 基于Egocentric视频的目标跟踪引起研究人员的DOI: 10.11992/tis.201709003 网络出版地址: http://kns.cnki.net/kcms/detail/23.1538.tp.20180417.1832.013.html 一种基于 Multi-Egocentric 视频运动轨迹重建的 多目标跟踪算法 欧伟奇1,2，尹辉1,2，许宏丽1,2，刘志浩1,2 （1. 北京交通大学 计算机与信息技术学院，北京 100044; 2. 北京交通大学 交通数据分析与挖掘北京市重点实 验室，北京 100044） 摘 要：Egocentric 视频具有目标运动剧烈、遮挡频繁、目标尺度差异明显及视角时变性强的特点，给目标跟踪 任务造成了极大的困难。本文从重建不同视角 Egocentric 视频中各目标的运动轨迹出发，提出一种基于 Multi￾Egocentric 视频运动轨迹重建的多目标跟踪算法，该方法基于多视角同步帧之间的单应性约束解决目标遮挡和 丢失问题，然后根据多视角目标空间位置约束关系通过轨迹重建进一步优化目标定位，并采用卡尔曼滤波构建 目标运动模型优化目标运动轨迹，在 BJMOT、EPLF-campus4 数据集上的对比实验验证了本文算法在解决 Multi-Egocentric 视频多目标跟踪轨迹不连续问题的有效性。 关键词：Multi-egocentric 视频；轨迹重建；多目标跟踪；单应性约束；对极几何约束；空间重构；卡尔曼滤波；运动 模型 中图分类号：TP391.4 文献标志码：A 文章编号：1673−4785(2019)02−0246−08 中文引用格式：欧伟奇, 尹辉, 许宏丽, 等. 一种基于 Multi-Egocentric 视频运动轨迹重建的多目标跟踪算法[J]. 智能系统学报, 2019, 14(2): 246–253. 英文引用格式：OU Weiqi, YIN Hui, XU Hongli, et al. A multi-object tracking algorithm based on trajectory reconstruction on multi-egocentric video[J]. CAAI transactions on intelligent systems, 2019, 14(2): 246–253. A multi-object tracking algorithm based on trajectory reconstruction on multi-egocentric video OU Weiqi1,2 ，YIN Hui1,2 ，XU Hongli1,2 ，LIU Zhihao1,2 (1. Department of Computer and Information Technology, Beijing Jiaotong University, Beijing 100044, China; 2. Beijing Key Lab of Transportation Data Analysis and Mining, Beijing Jiaotong University, Beijing 100044, China) Abstract: In egocentric video, objects have the characteristics of violent motion, frequent occlusion, so it brings much trouble to carrying out the tracking task. In this paper, we propose a multi-object tracking algorithm based on the mo￾tion trajectory reconstruction of multi-egocentric video from different visual angles egocentric videos. First, this method is based on the homographic constraint of multi-view synch frames to fix position of occluded and missing object. Second, using the relative position constraint relation of multi-angle target, the trajectory is reconstructed to locate the target position. Meanwhile, the trajectory of the object is optimized by constructing the motion model of object. Then the continuous trajectory of the object is obtained and the problem of the discontinuity trajectory in multi-object track￾ing is resolved. In the end, the performance of proposed method has been verified by using the multi-view video dataset which is created by us. Keywords: Multi-Egocentric video; trajectory reconstruction; multi-object tracking; homographic constraint; epipolar geometry constraint; spatial reconstruction; Kalman filter; motion model 目标跟踪是计算机视觉重要研究领域之一， 在智能交通、运动分析、行为识别、人机交互[1]等 方面具有广泛应用。随着可穿戴式相机的普及， 基于 Egocentric 视频的目标跟踪引起研究人员的 收稿日期：2017−09−05. 网络出版日期：2018−04−18. 基金项目：国家自然科学基金项目 (61472029，61473031)；科技 部国家重点研发计划项目 (2017YFB1201104，2016YFB 1200100)；中央高校基本科研业务费专项资金项目 (2016JBZ005). 通信作者：尹辉. E-mail：hyin@bjtu.edu.cn. 第 14 卷第 2 期 智 能 系 统 学 报 Vol.14 No.2 2019 年 3 月 CAAI Transactions on Intelligent Systems Mar. 2019