自适应交互式融合的视觉跟踪

发布时间：2018-07-01 19:04

本文选题：视觉跟踪 + 最小均方二乘误差　；参考：《光学精密工程》2017年09期

【摘要】：针对基于传统融合机制的联合跟踪器在复杂环境下鲁棒性不足的缺陷,提出一种在交互式多模型粒子滤波框架下传递概率矩阵可在线更新的自适应融合跟踪器。首先,在贝叶斯理论框架下,基于最小二乘误差估计法得到传递概率矩阵迭代更新方程;然后,利用数值积分法获得迭代更新方程的数值解;最后,结合重采样技术实现不同子跟踪器之间先验状态分布的自适应交互,以确保传递权值较大粒子对应的目标状态。在复杂环境下进行了的跟踪实验,结果验证了本文提出的自适应交互式融合机制增加了对粒子先验状态的校正功能,有效避免了因误差积累导致的"跟踪漂移"问题,使联合跟踪器的鲁棒性明显优于单一跟踪器或基于其它融合机制的联合跟踪器。
[Abstract]:An adaptive fusion tracker based on traditional fusion mechanism is proposed, which can update the transfer probability matrix online in the framework of interactive multi-model particle filter. Firstly, under the framework of Bayesian theory, the transfer probability matrix iterative updating equation is obtained based on the least square error estimation method. Then, numerical integration method is used to obtain the numerical solution of the iterative update equation. The resampling technique is used to realize the adaptive interaction of the prior state distribution between different sub-trackers to ensure the target state corresponding to the larger particle weight. The tracking experiments in complex environment show that the adaptive interactive fusion mechanism proposed in this paper increases the correction function of the priori state of particles, and effectively avoids the problem of "tracking drift" caused by the accumulation of errors. The robustness of the federated tracker is obviously superior to that of the single tracker or the federated tracker based on other fusion mechanisms.
【作者单位】：阜阳师范学院计算机与信息工程学院;安徽大学计算机科学与技术学院;
【基金】：安徽省高校优秀青年骨干人才支持计划资助项目(No.gxfx2017072) 阜阳师范学院青年人才基金重点资助项目(No.rcxm201706) 安徽省自然科学基金资助项目(No.1708085MF155)
【分类号】：TP391.41

【相似文献】