运动目标分割和跟踪.docVIP

华 中 科 技 大 学 硕 士 学 位 论 文  摘  要  本文重点研究视频序列图像中运动目标的分割与跟踪算法。视频序列分析通常 具有很强的任务依赖性，往往是针对不同的应用选用相应的算法。为解决视频序列 中对运动目标的分割与跟踪的问题,本文根据所处理的视频图像的特点,提出了视频 运动目标分割与跟踪系统的理论模型，为视频运动目标的描述与处理建立了相应的 理论基础和数学描述。 为了从动态背景中分割出运动目标，必须先对背景的运动进行补偿。本文中， 假设相邻两帧的背景变化可看作一个六参数的仿射变换，通过特征块的互相关匹配 法获得仿射变换参数，使背景对齐。结合帧间差分技术获得运动目标的大体区域和 少数噪声。然后进行行和列的区域填充，用面积阈值去掉噪声，再结合 canny 边缘 图像分割出完整的运动目标区域。 针对目标跟踪，本文讨论了卡尔曼预测的方法。根据系统的要求，确定了卡尔 曼预测的参数，并由此确定了具有预测稳定性的系统状态向量、状态转移矩阵和观 测矩阵，继而跟踪出运动目标的轨迹。 关键词：背景补偿；互相关匹配；差分法；Canny 边缘检测；卡尔曼滤波器 I 华 中 科 技 大 学 硕 士 学 位 论 文 Abstract The thesis is focused on the study of algorithm for moving target segmentation and tracking in video sequence images. Video sequence analysis is task-dependent. According to different applications, the corresponding algorithm will be selected. In this dissertation the theoretical model for the segmentation and tracking of the video moving object has been set up, which offers the mathematic description about the video moving object. In order to segment the moving objects from moving video background it is necessary to compensate the moving background. In this paper, we assume that the apparent background motion between two consecutive image frames can be approximated by affine transformation. In order to register the static background, we estimate affine transformation parameters using correlation-based block matching technique. Combine with inter-frame differential technique to get moving object region and a little noise. Then fill the line and row in the region, remove the noise through the area threshold. Combine with canny edge image to segmentation perfect moving object region. Method of Kalman filtering is discussed in terms of tracking moving object. Parameters of Kalman forecast are designed according to requirement of system and subsequently, system state vector, state transfer matrix and observation matrix are determined, and then tracking the trace of the moving object. Keywords: background compensation, cross-correlation match, difference, Can