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

多张面孔检测与跟踪

1.2 需要将安装包放在matlab 的当前路径,双击 usbwebcams.mlpkginstall

多张面孔检测与跟踪

1.3 点击该文件根据提示,进行安装

多张面孔检测与跟踪

多张面孔检测与跟踪

 

多张面孔检测与跟踪

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

多张面孔检测与跟踪

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) 

多张面孔检测与跟踪

 

欢迎加群 

多张面孔检测与跟踪

群资料 

多张面孔检测与跟踪

多张面孔检测与跟踪

多张面孔检测与跟踪

多张面孔检测与跟踪

多张面孔检测与跟踪

多张面孔检测与跟踪

多张面孔检测与跟踪

 

 

上一篇:2021-06-02


下一篇:基于opencv,openpose及caffe模型库实现手势识别