基于三维坐标运动目标跟踪.PDFVIP

第 45 卷第 2 期 应 用 科 技 Vol.45 No.2 2018 年 4 月 Applied Science and Technology Apr. 2018 DOI: 10.11991/yykj.201612007 网络出版地址：/kcms/detail/23.1191.U1549.008.html 基于三维坐标的运动目标跟踪 陈春雨，阳秋光，李东方 哈尔滨工程大学 信息与通信工程学院，黑龙江 哈尔滨 150001 摘 要：基于二维坐标的多运动目标跟踪，在跟踪过程中由于目标相互遮挡，算法无法分清各个运动目标，导致跟踪目标失 败。而三维坐标具有深度信息，利用目标遮挡前后坐标的不突变性能很好地分清各个目标，为此提出基于三维坐标的运动 目标跟踪方法。首先，采用背景差法进行目标检测；其次使用sift 算法对目标特征提取，运用极线约束对目标特征点进行立 体匹配以及三维重构并使用模糊C均值聚类算法(FCM)，确定运动目标中心三维坐标；最后结合Kalman 滤波实现目标跟 踪。实验和分析结果表明，算法能够较好地适应目标遮挡下的跟踪，具有良好的准确性、鲁棒性。 关键词：背景差法；sift 算法；三维重构；FCM 聚类；Kalman 滤波；三维坐标；二维坐标；立体匹配 中图分类号：TN957 文献标志码：A 文章编号：1009−671X(2018)02−0023−06 Moving objective tracking based on 3D coordinates CHEN Chunyu, YANG Qiuguang, LI Dongfang College of Information and Communication Engineering, Harbin Engineering University, Harbin 150001, China Abstract: As for multi-objective tracking based on two-dimensional coordinates, because of mutual objective obstruction, the algorithm is hard to distinguish from moving objectives and the objective tracking may fail; while three- dimensional coordinates have depth information, the moving objectives can be clearly identified due to the non-mutation of coordinates before and after objective obstruction, therefore, the paper proposed a new multi-objective tracking algorithm based on three-dimensional coordinates. Firstly, the background subtraction method is used to detect objectives; secondly, sift algorithm is used to extract features of objectives, the epipolar constraint is utilized to carry out stereo matching and the three-dimensional reconstruction for the objective feature points, in addition, fuzzy