函数obj = setupSystemObjects ()%初始化视频I / O%为阅读视频从一个文件中创建对象,跟踪%每一帧中的对象,和玩视频。%创建一个电子阅读器。obj。读者= VideoReader (“atrium.mp4”);%创建两个视频播放器,一个显示视频,%和一个显示前台面具。obj。maskPlayer = vision.VideoPlayer (“位置”,740,400,700,400);obj。放像机= vision.VideoPlayer (“位置”,400,700,400);%为前景检测和blob分析创建系统对象%前台探测器是用来段移动对象%的背景。它输出二进制掩模,像素值% 1的值对应于前景和0对应%的背景。obj.detector = vision.ForegroundDetector (“NumGaussians”3,…“NumTrainingFrames”现年40岁的“MinimumBackgroundRatio”,0.7);%连接组的前景像素可能对应于运动%的对象。blob分析系统对象用于发现这些组织%(称为“blob”或“连接组件”),并计算他们%的特征,如面积、质心和边界框。obj。blobAnalyser = vision.BlobAnalysis (“BoundingBoxOutputPort”,真的,…“AreaOutputPort”,真的,“CentroidOutputPort”,真的,…“MinimumBlobArea”,400);结束

函数跟踪= initializeTracks ()%的创建一个空数组跟踪=结构(…“id”{},…“bbox”{},…“kalmanFilter”{},…“年龄”{},…“totalVisibleCount”{},…“consecutiveInvisibleCount”,{});结束

函数[重心,bboxes,面具]= detectObjects(框架)%检测前景。掩码= obj.detector.step(框架);%应用形态学操作来消除噪音和填补漏洞。掩码= imopen(面具,strel (“矩形”[3 3]));掩码= imclose(面具,strel (“矩形”[15、15]));= imfill(面具,面具“黑洞”);%执行团分析找到连接组件。[~,重心,bboxes] = obj.blobAnalyser.step(面具);结束

函数predictNewLocationsOfTracks ()为i = 1:长度(跟踪)bbox =跟踪(我).bbox;%预测的当前位置跟踪。predictedCentroid =预测(跟踪(i) .kalmanFilter);%将边界框,其中心%预计的位置。predictedCentroid = int32 (predictedCentroid)——bbox (3:4) / 2;跟踪(i)。bbox= [predictedCentroid, bbox(3:4)];结束结束

函数(作业、unassignedTracks unassignedDetections] =…detectionToTrackAssignment () nTracks =长度(跟踪);nDetections =大小(质心,1);%计算的成本分配每个检测跟踪。成本= 0 (nTracks nDetections);为i = 1: nTracks成本(我,:)=距离(跟踪(i)。kalmanFilter、质心);结束%解决分配问题。costOfNonAssignment = 20;(作业、unassignedTracks unassignedDetections] =…assignDetectionsToTracks(成本、costOfNonAssignment);结束

函数updateAssignedTracks () numAssignedTracks =大小(作业,1);为i = 1: numAssignedTracks trackIdx =作业(我,1);detectionIdx =作业(我,2);质心=质心(detectionIdx:);bbox = bboxes (detectionIdx:);%的正确估计对象的位置%使用新的检测。正确的(跟踪(trackIdx)。kalmanFilter、质心);%替代预测与检测到的边界框%边界框。跟踪(trackIdx)。bbox = bbox;%更新追踪的年龄。跟踪(trackIdx)。年龄= tracks(trackIdx).age + 1;%更新的可见性。跟踪(trackIdx)。totalVisibleCount=…跟踪(trackIdx)。totalVisibleCount+ 1; tracks(trackIdx).consecutiveInvisibleCount = 0;结束结束

函数updateUnassignedTracks ()为i = 1:长度(unassignedTracks)印第安纳= unassignedTracks(我);跟踪(印第安纳州)。年龄= tracks(ind).age + 1; tracks(ind).consecutiveInvisibleCount =…跟踪(印第安纳州)。consecutiveInvisibleCount + 1;结束结束

函数deleteLostTracks ()如果isempty(跟踪)返回;结束invisibleForTooLong = 20;ageThreshold = 8;%计算分数的跟踪的年龄这是可见的。年龄=(跟踪(:).age);totalVisibleCounts =(跟踪(:).totalVisibleCount);可见性= totalVisibleCounts。/年龄;%找到《迷失》的指数跟踪。lostInds =(年龄< ageThreshold & < 0.6) |的可见性…(跟踪(:)。consecutiveInvisibleCount] > = invisibleForTooLong;%删除失去了踪迹。跟踪=跟踪(~ lostInds);结束

函数createNewTracks()质心=质心(unassignedDetections:);bboxes = bboxes (unassignedDetections:);为i = 1:尺寸(质心,1)重心=质心(我:);bbox = bboxes(我);%创建一个卡尔曼滤波器对象。kalmanFilter = configureKalmanFilter (“ConstantVelocity”,…重心,200,,100,25,100);%创建一个新的轨道。newTrack =结构(…“id”nextId,…“bbox”bbox,…“kalmanFilter”kalmanFilter,…“年龄”,1…“totalVisibleCount”,1…“consecutiveInvisibleCount”,0);%将它添加到数组的痕迹。跟踪(结束+ 1)= newTrack;%增加下一个id。nextId = nextId + 1;结束结束

函数displayTrackingResults ()%转换框架和uint8 RGB的面具。= im2uint8帧(帧);掩码= uint8 (repmat(面具,(1,1,3)))* 255;minVisibleCount = 8;如果~ isempty(跟踪)%的检测往往导致短暂的痕迹。%只显示跟踪已超过可见%最小的帧数。reliableTrackInds =…(跟踪(:)。totalVisibleCount] > minVisibleCount;reliableTracks =跟踪(reliableTrackInds);%显示对象。如果一个对象没有被发现%在本帧,显示其预测的边界框。如果~ isempty (reliableTracks)%得到边界框。reliableTracks.bbox bboxes =猫(1);%得到id。id = int32 ([reliableTracks (:) .id]);%为对象创建标签的指示%,我们预测,而不是实际的显示%的位置。标签= cellstr (int2str (ids));predictedTrackInds =…[reliableTracks (:)。consecutiveInvisibleCount] > 0;isPredicted =细胞(大小(标签));isPredicted (predictedTrackInds) = {“预测”};标签= strcat(标签、isPredicted);%画出对象的框架。= insertObjectAnnotation帧(帧,“矩形”,…bboxes、标签);%画出对象的面具。= insertObjectAnnotation(面具,面具“矩形”,…bboxes、标签);结束结束%显示面具和框架。obj.maskPlayer.step(面具);obj.videoPlayer.step(框架);结束

结束