计算机技术与发展2018，Vol.28Issue(6)：93-96,101,5.DOI:10.3969/j.issn.1673-629X.2018.06.021
基于模型融合和特征关联的视频目标跟踪算法
Video Target Tracking Algorithm Based on Model Fusion and Feature Association
摘要
Abstract
In order to solve the problems of background noise and target occlusion for target tracking in video,we present a multi-target tracking algorithm in video based on model fusion and feature association. It first improves the symmetric interframe difference method, namely,using the continuous four-frame images to execute interframe difference and performing the"or" operation to extract the target contours. Then,applying the results to perform the "and" operation with the contours extracted from sliding average background differ-ence,we can get the centroid positions of moving targets after corrosion and expansion operations in morphological. Subsequently,we fuse pyramid optical flow method and Kalman filter to predict the centroid positions of targets at the next moment and employ Hungarian algo-rithm to calculate optimal matches for feature correlation,while removing the parts of tracker that don' t meet the requirements and crea-ting tracking units for unassigned detections. Finally,we correct the parameters of Kalman filter and draw the trajectories of the moving targets. The experiments show that the proposed algorithm can extract the contours of moving targets accurately,solve the tracking failure caused by mutual occlusion between targets,and improve the accuracy and robustness of multi-target tracking effectively.关键词
多目标跟踪/对称帧间差分/金字塔光流/卡尔曼滤波Key words
multi-target tracking/symmetric interframe difference/pyramid optical flow/Kalman filter分类
信息技术与安全科学引用本文复制引用
季露,陈志,岳文静..基于模型融合和特征关联的视频目标跟踪算法[J].计算机技术与发展,2018,28(6):93-96,101,5.基金项目
国家自然科学基金(61501253) （61501253）
中国博士后科学基金项目( 2013M531393 ) （ 2013M531393 ）
江苏省基础研究计划(自然科学基金)项目(BK20151506) （自然科学基金）
江苏省"六大人才高峰"第十一批高层次人才选拔培养资助项目(XXRJ-009) （XXRJ-009）
江苏省重点研发计划(社会发展)项目(BE2016778) （社会发展）
南京邮电大学科研项目(NY217054) （NY217054）
3S杯全国大学生物联网技术与应用"三创"大赛立项项目(17B077) （17B077）
