基于随机有限集理论的复杂背景视频多目标跟踪研究

项目名称： 基于随机有限集理论的复杂背景视频多目标跟踪研究

项目编号： No.61573281

项目类型： 面上项目

立项/批准年度： 2016

项目学科： 自动化技术、计算机技术

项目作者： 鲁晓锋

作者单位： 西安理工大学

项目金额： 16万元

中文摘要： 传统的多目标跟踪方法受制于数据关联算法，在复杂的多目标跟踪环境下通常会引发“组合爆炸”和NP-hard难题。基于随机集理论的滤波跟踪方法具有系统、严格的数学基础，有效避免多目标跟踪中复杂的数据关联问题。对于多目标跟踪，通过对随机集滤波算法的分析，针对目标漏检问题，提出基于高斯分量的权值动态再分配机制的势估计概率假设密度滤波；针对目标数过估和量测更新信息被忽略问题，提出基于目标权值合并机制的势估计多贝努利滤波。本项目通过提出对应的解决机制，从而完善随机集理论的滤波算法，为多目标跟踪的研究与应用提供了新的思路和理论依据。

中文关键词： 多目标跟踪；视频目标跟踪；随机有限集；；

英文摘要： The traditional multi-target tracking methods are limited by the complicated data association. These bring combination explosion and N-P hard problem in complex situations. The filter algorithms based on random finite set theory provide systematic and rigorous mathematical foundation for multi-target tracking method, and avoid using data association technique.In the multi target tracking stage, the project analyses the characteristics of the filter algorithm based on random finite set theory, proposes the improved cardinalized probability hypothesis density (CPHD) filter based on the dynamic redistribution mechanism of the weights of Gaussian components to solve the missed detections. To solve over-estimates the cardinality and reduction measurement information, the improved cardinality balanced multi-target multiple Bermoulli (CBMeMBer) filter algorithm based on the combined mechanism of target weights is proposed, improves the effectiveness of the filter algorithms.This project puts forward the solving mechanisms corresponding to these filter algorithms, so as to improves the filter algorithms of random finite set theory, and provides a new idea and theory basis for the research and application of multi target tracking.

英文关键词： Multi-target tracking；Visual target tracking；Random finite set；；