使用实时视频采集人脸检测和跟踪

%创建面对探测器对象。faceDetector = vision.CascadeObjectDetector ();%创建重点跟踪对象。pointTracker = vision.PointTracker (“MaxBidirectionalError”2);%创建远程对象。凸轮=摄像头();%获取一帧大小。videoFrame =快照(cam);frameSize =大小(videoFrame);%创建视频播放器对象。放像机= vision.VideoPlayer (“位置”100年[100 [frameSize (2), frameSize (1)] + 30]);

runLoop = true;numPts = 0;frameCount = 0;而runLoop & & frameCount < 400%得到下一帧。videoFrame =快照(cam);videoFrameGray = im2gray (videoFrame);frameCount = frameCount + 1;如果numPts < 10%检测模式。bbox = faceDetector.step (videoFrameGray);如果~ isempty (bbox)%找到角点检测区域内。点= detectMinEigenFeatures (videoFrameGray,“投资回报”bbox (:));%初始化跟踪。xyPoints = points.Location;numPts =大小(xyPoints, 1);释放(pointTracker);初始化(pointTracker xyPoints videoFrameGray);%保存点的一个副本。oldPoints = xyPoints;%将矩形表示为(x, y, w h)成一个% M-by-2 (x, y)坐标矩阵的四个角。这%是需要能够改变边界框显示%的方向面对。:bboxPoints = bbox2points (bbox (1));%将盒子角落转换成(x1 y1 x2 y2 x3 y3 x4 y4)% insertShape所需的格式。bboxPolygon =重塑(bboxPoints ', 1, []);%显示一个边界框周围发现的脸。videoFrame = insertShape (videoFrame,“多边形”bboxPolygon,“线宽”3);%显示检测到的角落。videoFrame = insertMarker (videoFrame xyPoints,“+”,“颜色”,“白色”);结束其他的%跟踪模式。[xyPoints, isFound] = (pointTracker videoFrameGray)步;visiblePoints = xyPoints (isFound:);oldInliers = oldPoints (isFound:);numPts =大小(visiblePoints, 1);如果numPts > = 10%估计老点之间的几何变换%和新观点。[xform, inlierIdx] = estgeotform2d (…oldInliers visiblePoints,“相似”,“MaxDistance”4);oldInliers = oldInliers (inlierIdx:);visiblePoints = visiblePoints (inlierIdx:);%转换应用于边界框。bboxPoints = transformPointsForward (xform, bboxPoints);%将盒子角落转换成(x1 y1 x2 y2 x3 y3 x4 y4)% insertShape所需的格式。bboxPolygon =重塑(bboxPoints ', 1, []);%显示一个边界框的脸被跟踪。videoFrame = insertShape (videoFrame,“多边形”bboxPolygon,“线宽”3);%显示跟踪点。videoFrame = insertMarker (videoFrame visiblePoints,“+”,“颜色”,“白色”);%重置点。oldPoints = visiblePoints;选点(pointTracker oldPoints);结束结束%显示注释视频帧使用视频播放器对象。步骤(放像机、videoFrame);%检查是否视频播放器窗口已经关闭。runLoop = isOpen(放像机);结束%清理。清晰的凸轮;释放(放像机);释放(pointTracker);释放(faceDetector);