基于时序空间关系的目标跟踪及遮挡识别研究

项目名称： 基于时序空间关系的目标跟踪及遮挡识别研究

项目编号： No.61502026

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 计算机科学学科

项目作者： 滕竹

作者单位： 北京交通大学

项目金额： 21万元

中文摘要： 本研究针对目标跟踪问题中如何描述目标的外观模型，如何分析目标前后帧之间的关系，以及如何识别和标注视频中遮挡等三个问题，从不同角度展开分析，以三个研究内容为切入点，提出两种目标跟踪的研究思路。第一个研究内容是从跟踪问题中特有的时序空间特征的角度建立被跟踪对象的外观模型，在粒子滤波的大框架下，重点研讨如何通过历史帧重构当前帧目标。第二个研究内容是将目标栅格化并以子块的形式表达，着重研究目标子块间的时序相关性和空间有序性并考虑子块间的约束关系、竞争机制等以通过优化方法确定目标的位置。最后，针对两种思路共同存在的遮挡问题，利用前帧的目标及邻域背景重构当前帧目标以及当前帧与前帧子块的反匹配关系识别和标注遮挡位置。本项目的实施将进一步丰富和完善视觉跟踪研究领域的新内容，特别是遮挡识别的提出，不仅使跟踪器的性能更稳定，同时为涉及到遮挡问题的其他领域提供必要的基础，有望取得创新成果，具有重要的科学意义。

中文关键词： 目标跟踪；遮挡识别；时序空间关系

英文摘要： This research concentrates on three problems in object tracking: how to develop an appearance model to present the tracking object; how to describe the context relationships of the objects between frames; and how to recognize and label occlusions in the video sequences. These problems are analyzed from two different perspectives and two proposals for object tracking are presented through three research points. The first research point focuses on a particular appearance model for tracking, which is developed by exploring the temporal and spatial relationships of targets, and a tracking algorithm is proposed based on particle filters. The second research point aims at employing rasterizing cells to represent targets and establishing a tracking algorithm under a framework of optimization. The last research core is the occlusion recognition that both two proposals face. We propose to identify occlusions by reconstructing the target of the current frame via the target and its neighborhood in the previous frame and by advancing a contra-matching relationship between cells from adjacent frames. The implementation of this research will further enrich and advance the research of the tracking field, especially the proposal of occlusion recognition, which not only makes the tracker more robust, but also provides a basis for other areas confronting the problem of occlusion. It is expected to yield some innovative outcomes with significant scientific benefits.

英文关键词： Object Tracking;Occlusion Recognition;temporal and spatial relationships