计算机技术与发展2018，Vol.28Issue(5)：27-31,5.DOI:10.3969/j.issn.1673-629X.2018.05.007
基于弱特征重识别的多目标长效摘要
Long Term Abstraction of Multi-object Based on Weak Feature Re-identification
摘要
Abstract
There are a large number of occluded target in the real video surveillance system.The target can't be re-identified when occlu-sion occurs,then the trajectory of the same target is divided into segmentations and can't be associated.For this,we use weak features of the target to recognize itself again after the occurrence of occlusion and the appearance of target.Firstly,the predicted position of the tar-get is derived by Kalman filtering.Then,the joint probability data association algorithm is introduced to calculate the probability of the measurement belonging to the target.The measurement-to-target assignments means that each measurement is assigned to no more than one target,and each target is uniquely assigned to one measurement.Finally,we update the target state with the measurement which has the maximum probability.In the case that we can't get the accurate state of the target because of occlusion,consider the new target ap-peared in the middle of the region as the target which is lost before,so we extract the target's weak features for future use when it is stab-ly tracked.After the occlusion we look for the new target's feature similarity by semicircle search which is formed about 90 degrees left or right of the original moving direction of the target.With high similarity it is considered as the object match to solve the occlusion.The experiment shows that in comparison with the original algorithm,the presented algorithm reduces the total number of abstracts obtained and the probability of target abstracts being segmented,and increases the correct number of summaries,which improves the robustness of the tracking system with a long-term summary of the target.关键词
多目标跟踪/弱特征/卡尔曼滤波/联合概率数据关联算法/重识别Key words
multi-target tracking/weak features/Kalman filtering/joint probabilistic data association algorithm/re-identification分类
信息技术与安全科学引用本文复制引用
石亚玲,刘正熙,熊运余,李征..基于弱特征重识别的多目标长效摘要[J].计算机技术与发展,2018,28(5):27-31,5.基金项目
国家自然科学基金面上项目(61471250) （61471250）
