摘要
针对实时目标跟踪会产生跟踪不稳定、易漂移、被遮挡就丢失的问题,提出改进的多样本跟踪算法。在压缩传感实时跟踪中,通过增加随机测量矩阵产生新的压缩感知特征,融合多个正负样本。结合boosting学习方法更新特征权值并改进置信图估计,解决目标漂移和丢失问题。实验结果表明,该方法在目标运动、纹理和环境显著变化以及被部分遮挡的情况下,跟踪的鲁棒性依旧很高,能达到稳定、实时的目标跟踪。
Aiming at the problems of unstable tracking , easy to drift , obscured loss , which are produced in real-time target track-ing, we propose an improved tracking algorithm for multiple instances .In the compressed sensing and real-time tracking, by adding random measurement matrix to produce new features , multiple positive and negative instances are integrated .By combi-ning with the boosting learning method to update the feature weights and improve the confidence map estimation , we solve the problems of target drift and loss .Experimental results show that the proposed algorithm achieves better robustness and stable real -time tracking when the target moves quickly , or in conditions that the textures and lightings change seriously , as well as it is par-tially covered .
出处
《计算机与现代化》
2014年第7期58-62,67,共6页
Computer and Modernization
基金
江苏省六大人才高峰项目(DZXX-028)
江南大学教师卓越工程项目(JGC2013145)
关键词
目标跟踪
压缩传感
多样本
实时跟踪
漂移
目标遮挡
target tracking
compressive sensing
multiple instance
real-time tracking
drift
target occlusion
