摘要
基于特征融合的目标跟踪中,目标的特征由于某些干扰导致准确度较低。基于贝叶斯框架的特征融合算法进行目标跟踪时,不能达到最佳的跟踪效果。为了更好地融合目标的不同特征来实现不同场景的视频中的目标匹配,提出了一种新的特征融合算法。该算法在特征融合过程中引入特征匹配的置信水平信息作为正确目标选择的依据,然后采用不同于粒子滤波的迭代算法确定匹配的正确目标。实验结果表明,对于参与融合的特征之间有相对高的性能和特征间差异性较大的情况,新的融合算法比基于贝叶斯框架的融合算法在大部分情况下具备更高的匹配准确性。
target tracking based on features fusion , features for some interference lead to low accuracy. Feature fusion algorithm for target tracking based on the bayesian framework, cannot achieve the best tracking performance. In order to fusion better the different features of target to realize object matching in different scene of the video, proposed a new feature fusion algorithm. The algorithm in feature fusion process introduces feature matching confidence level information as the basis of selection for the right target, then uses the different iterative algorithm from particle filtering to determine the correct target matching. The experimental results show： among the features of fusion ,they have relatively high performance and larger difference ,the new fusion algorithm has higher matching accuracy than the fusion algorithm based on bayesian framework in most cases.
出处
《电子设计工程》
2013年第12期102-104,共3页
Electronic Design Engineering
基金
河南省科技发展计划项目(1321022210479)
关键词
目标匹配
特征融合
贝叶斯框架
粒子滤波
置信度
target matching
feature fusion
bayesian framework
particle filter
confidence
