多张面孔检测与跟踪
多张面孔检测与跟踪
内容
-
环境配置
-
运行代码
1.1 访问:Webcam Support Package。击右下角位置的download,下载Webcam安装包
1.2 需要将安装包放在matlab 的当前路径,双击 usbwebcams.mlpkginstall
1.3 点击该文件根据提示,进行安装
在安装完成结束后,在命令行输入:webcam后出现下图属性情况,则说明安装成功
运行代码
%% detectAndTrackFaces
% Automatically detects and tracks multiple faces in a webcam-acquired
% video stream.
%
% Copyright 2013-2014 The MathWorks, Inc
clear classes;
%% Instantiate video device, face detector, and KLT object tracker
vidObj = webcam;
faceDetector = vision.CascadeObjectDetector(); % Finds faces by default
tracker = MultiObjectTrackerKLT;
%% Get a frame for frame-size information
frame = snapshot(vidObj);
frameSize = size(frame);
%% Create a video player instance
videoPlayer = vision.VideoPlayer('Position',[200 100 fliplr(frameSize(1:2)+30)]);
%% Iterate until we have successfully detected a face
bboxes = [];
while isempty(bboxes)
framergb = snapshot(vidObj);
frame = rgb2gray(framergb);
bboxes = faceDetector.step(frame);
end
tracker.addDetections(frame, bboxes);
%% And loop until the player is closed
frameNumber = 0;
keepRunning = true;
disp('Press Ctrl-C to exit...');
while keepRunning
framergb = snapshot(vidObj);
frame = rgb2gray(framergb);
if mod(frameNumber, 10) == 0
% (Re)detect faces.
%
% NOTE: face detection is more expensive than imresize; we can
% speed up the implementation by reacquiring faces using a
% downsampled frame:
% bboxes = faceDetector.step(frame);
bboxes = 2 * faceDetector.step(imresize(frame, 0.5));
if ~isempty(bboxes)
tracker.addDetections(frame, bboxes);
end
else
% Track faces
tracker.track(frame);
end
% Display bounding boxes and tracked points.
displayFrame = insertObjectAnnotation(framergb, 'rectangle',...
tracker.Bboxes, tracker.BoxIds);
displayFrame = insertMarker(displayFrame, tracker.Points);
videoPlayer.step(displayFrame);
frameNumber = frameNumber + 1;
end
%% Clean up
release(videoPlayer);
以下函数命名为:MultiObjectTrackerKLT.m
classdef MultiObjectTrackerKLT < handle
properties
% PointTracker A vision.PointTracker object
PointTracker;
% Bboxes M-by-4 matrix of [x y w h] object bounding boxes
Bboxes = [];
% BoxIds M-by-1 array containing ids associated with each bounding box
BoxIds = [];
% Points M-by-2 matrix containing tracked points from all objects
Points = [];
% PointIds M-by-1 array containing object id associated with each
% point. This array keeps track of which point belongs to which object.
PointIds = [];
% NextId The next new object will have this id.
NextId = 1;
% BoxScores M-by-1 array. Low box score means that we probably lost the object.
BoxScores = [];
end
methods
%------------------------------------------------------------------
function this = MultiObjectTrackerKLT()
% Constructor
this.PointTracker = ...
vision.PointTracker('MaxBidirectionalError', 2);
end
%------------------------------------------------------------------
function addDetections(this, I, bboxes)
% addDetections Add detected bounding boxes.
% addDetections(tracker, I, bboxes) adds detected bounding boxes.
% tracker is the MultiObjectTrackerKLT object, I is the current
% frame, and bboxes is an M-by-4 array of [x y w h] bounding boxes.
% This method determines whether a detection belongs to an existing
% object, or whether it is a brand new object.
for i = 1:size(bboxes, 1)
% Determine if the detection belongs to one of the existing
% objects.
boxIdx = this.findMatchingBox(bboxes(i, :));
if isempty(boxIdx)
% This is a brand new object.
this.Bboxes = [this.Bboxes; bboxes(i, :)];
points = detectMinEigenFeatures(I, 'ROI', bboxes(i, :));
points = points.Location;
this.BoxIds(end+1) = this.NextId;
idx = ones(size(points, 1), 1) * this.NextId;
this.PointIds = [this.PointIds; idx];
this.NextId = this.NextId + 1;
this.Points = [this.Points; points];
this.BoxScores(end+1) = 1;
else % The object already exists.
% Delete the matched box
currentBoxScore = this.deleteBox(boxIdx);
% Replace with new box
this.Bboxes = [this.Bboxes; bboxes(i, :)];
% Re-detect the points. This is how we replace the
% points, which invariably get lost as we track.
points = detectMinEigenFeatures(I, 'ROI', bboxes(i, :));
points = points.Location;
this.BoxIds(end+1) = boxIdx;
idx = ones(size(points, 1), 1) * boxIdx;
this.PointIds = [this.PointIds; idx];
this.Points = [this.Points; points];
this.BoxScores(end+1) = currentBoxScore + 1;
end
end
% Determine which objects are no longer tracked.
minBoxScore = -2;
this.BoxScores(this.BoxScores < 3) = ...
this.BoxScores(this.BoxScores < 3) - 0.5;
boxesToRemoveIds = this.BoxIds(this.BoxScores < minBoxScore);
while ~isempty(boxesToRemoveIds)
this.deleteBox(boxesToRemoveIds(1));
boxesToRemoveIds = this.BoxIds(this.BoxScores < minBoxScore);
end
% Update the point tracker.
if this.PointTracker.isLocked()
this.PointTracker.setPoints(this.Points);
else
this.PointTracker.initialize(this.Points, I);
end
end
%------------------------------------------------------------------
function track(this, I)
% TRACK Track the objects.
% TRACK(tracker, I) tracks the objects into frame I. tracker is the
% MultiObjectTrackerKLT object, I is the current video frame. This
% method updates the points and the object bounding boxes.
[newPoints, isFound] = this.PointTracker.step(I);
this.Points = newPoints(isFound, :);
this.PointIds = this.PointIds(isFound);
generateNewBoxes(this);
if ~isempty(this.Points)
this.PointTracker.setPoints(this.Points);
end
end
end
methods(Access=private)
%------------------------------------------------------------------
function boxIdx = findMatchingBox(this, box)
% Determine which tracked object (if any) the new detection belongs to.
boxIdx = [];
for i = 1:size(this.Bboxes, 1)
area = rectint(this.Bboxes(i,:), box);
if area > 0.2 * this.Bboxes(i, 3) * this.Bboxes(i, 4)
boxIdx = this.BoxIds(i);
return;
end
end
end
%------------------------------------------------------------------
function currentScore = deleteBox(this, boxIdx)
% Delete object.
this.Bboxes(this.BoxIds == boxIdx, :) = [];
this.Points(this.PointIds == boxIdx, :) = [];
this.PointIds(this.PointIds == boxIdx) = [];
currentScore = this.BoxScores(this.BoxIds == boxIdx);
this.BoxScores(this.BoxIds == boxIdx) = [];
this.BoxIds(this.BoxIds == boxIdx) = [];
end
%------------------------------------------------------------------
function generateNewBoxes(this)
% Get bounding boxes for each object from tracked points.
oldBoxIds = this.BoxIds;
oldScores = this.BoxScores;
this.BoxIds = unique(this.PointIds);
numBoxes = numel(this.BoxIds);
this.Bboxes = zeros(numBoxes, 4);
this.BoxScores = zeros(numBoxes, 1);
for i = 1:numBoxes
points = this.Points(this.PointIds == this.BoxIds(i), :);
newBox = getBoundingBox(points);
this.Bboxes(i, :) = newBox;
this.BoxScores(i) = oldScores(oldBoxIds == this.BoxIds(i));
end
end
end
end
%--------------------------------------------------------------------------
function bbox = getBoundingBox(points)
x1 = min(points(:, 1));
y1 = min(points(:, 2));
x2 = max(points(:, 1));
y2 = max(points(:, 2));
bbox = [x1 y1 x2 - x1 y2 - y1];
end
运行效果图(因为是夜晚,只有我一人检测,所有没有多目标检测)
欢迎加群
群资料
posted on 2020-01-14 02:03 好玩的MATLAB 阅读(0) 评论(0) 编辑 收藏 举报 来源