搜索
西北工业大学本科毕业设计论文I目录摘要..............................................................................................................................1ABSTRACT...................................................................................................................2第一章绪论................................................................................................................41.1课题研究背景和意义.......................................................................................41.2国内外研究现状...............................................................................................51.3本文的具体结构安排.......................................................................................7第二章运动目标检测..................................................................................................82.1检测算法及概述...............................................................................................82.1.1连续帧间差分法.....................................................................................92.1.2背景去除法...........................................................................................112.1.3光流法...................................................................................................13第三章运动目标跟踪方法........................................................................................163.1引言.................................................................................................................163.2运动目标跟踪方法.........................................................................................163.2.1基于特征匹配的跟踪方法...................................................................163.2.2基于区域匹配的跟踪方法...................................................................173.2.3基于模型匹配的跟踪方法...................................................................183.3运动目标搜索算法.........................................................................................183.3.1绝对平衡搜索法....................................................................................183.4绝对平衡搜索法实验结果.............................................................................193.4.1归一化互相关搜索法...........................................................................213.5归一化互相关搜索法实验结果及分析.........................................................22第四章模板更新与轨迹预测..................................................................................264.1模板更新简述及策略.....................................................................................264.2轨迹预测.........................................................................................................284.2.1线性预测...............................................................................................294.2.2平方预测器...........................................................................................30西北工业大学本科毕业设计论文II4.3实验结果及分析：.........................................................................................31致谢............................................................................................................................36参考文献......................................................................................................................37毕业设计小结..............................................................................................................38西北工业大学本科毕业设计论文1摘要图像序列目标跟踪是计算机视觉中的经典问题，它是指在一组图像序列中，根据所需目标模型，实时确定图像中目标所在位置的过程。它最初吸引了军方的关注，逐渐被应用于电视制导炸弹、火控系统等军用备中。序列图像运动目标跟踪是通过对传感器拍摄到的图像序列进行分析，计算出目标在每帧图像上的位置。它是计算机视觉系统的核心，是一项融合了图像处理、模式识别、人工只能和自动控制等领域先进成果的高技术课题，在航天、监控、生物医学和机器人技术等多种领域都有广泛应用。因此，非常有必要研究运动目标的跟踪。本论文就图像的单目标跟踪问题，本文重点研究了帧间差分法和背景去除法等目标检测方法，研究了模板相关匹配跟踪算法主要是：最小均方误差函数(MES)，最小平均绝对差值函数(MAD)和最大匹配像素统计(MPC)的跟踪算法。在跟踪过程中，由于跟踪设备与目标的相对运动,视野中的目标可能出现大小、形状、姿态等变化,加上外界环境中的各种干扰,所要跟踪的目标和目标所在的场景都发生了变化,有可能丢失跟踪目标。为了保证跟踪的稳定性和正确性,需要对模板图像进行自适应更新。由于目标运动有一定得规律，可以采取轨迹预测以提高跟踪精度，本文采用了线性预测法。对比分析了相关匹配算法的跟踪精度和跟踪速度；对比不采用模板更新和模板跟新的跟踪进度和差别，实验表明，跟踪算法加上轨迹预测及模板跟新在很大程度上提高了跟踪帧数，提高了跟踪精度，具有一定的抗噪声性能。关键词：目标跟踪，目标检测，轨迹预测，模板更新西北工业大学本科毕业设计论文2ABSTRACTTargettracking,imagesequenceisaclassiccomputervisionproblems,itisdefinedasasetofimagesequences,inaccordancewithrequirementsofthetargetmodel,real-timeimagestodeterminethelocationofthetargetprocess.Itinitiallyattractedtheconcernofthemilitaryhasgraduallybeenappliedtotelevision-guidedbombs,firecontrolsystemsformilitarypreparation.Movingtargettrackingsensoristakenthroughtheimagesequenceanalysis,tocalculatethetargetimageineachframeposition.Itisthecoreofcomputervisionsystemisacombinationofimageprocessing,patternrecognition,artificialonlyandtheresultsofautomaticcontrolinareassuchasadvancedhigh-techissuesintheaerospace,control,biomedicalandroboticsfields,etc.Therearewidelyused.Thus,itisnecessarytostudythetrackingofmovingtargets.Inthispaper,theimageofthesingle-targettrackingproblem,researchthetargetdetectionmethodismainlybasedoninter-framedifferenceandbackgroundremovalmethodtodetectthetargetinpreparationfortargettracking.Templatematchingtrackingalgorithmis:thesmallestmeansquareerrorfunction(MES),thesmallestmeanabsolutedifferencefunction(MAD)andthemaximummatchingpix