TrafficSignDetection.m

%
% Template example for using on the validation set.
% 
 
function TrafficSignDetection(directory, pixel_method, window_method, decision_method)
    % TrafficSignDetection
    % Perform detection of Traffic signs on images. Detection is performed first at the pixel level
    % using a color segmentation. Then, using the color segmentation as a basis, the most likely window 
    % candidates to contain a traffic sign are selected using basic features (form factor, filling factor). 
    % Finally, a decision is taken on these windows using geometric heuristics (Hough) or template matching.
    %
    %    Parameter name      Value
    %    --------------      -----
    %    'directory'         directory where the test images to analize  (.jpg) reside
    %    'pixel_method'      Name of the color space: 'opp', 'normrgb', 'lab', 'hsv', etc. (Weeks 2-5)
    %    'window_method'     'SegmentationCCL' or 'SlidingWindow' (Weeks 3-5)
    %    'decision_method'   'GeometricHeuristics' or 'TemplateMatching' (Weeks 4-5)


    global CANONICAL_W;        CANONICAL_W = 64;
    global CANONICAL_H;        CANONICAL_H = 64;
    global SW_STRIDEX;         SW_STRIDEX = 8;
    global SW_STRIDEY;         SW_STRIDEY = 8;
    global SW_CANONICALW;      SW_CANONICALW = 32;
    global SW_ASPECTRATIO;     SW_ASPECTRATIO = 1;
    global SW_MINS;            SW_MINS = 1;
    global SW_MAXS;            SW_MAXS = 2.5;
    global SW_STRIDES;         SW_STRIDES = 1.2;


    % Load models
    %global circleTemplate;
    %global givewayTemplate;   
    %global stopTemplate;      
    %global rectangleTemplate; 
    %global triangleTemplate;  
    %
    %if strcmp(decision_method, 'TemplateMatching')
    %   circleTemplate    = load('TemplateCircles.mat');
    %   givewayTemplate   = load('TemplateGiveways.mat');
    %   stopTemplate      = load('TemplateStops.mat');
    %   rectangleTemplate = load('TemplateRectangles.mat');
    %   triangleTemplate  = load('TemplateTriangles.mat');
    %end

    % windowTP=0; windowFN=0; windowFP=0; % (Needed after Week 3)
    pixelTP=0; pixelFN=0; pixelFP=0; pixelTN=0;
    
    files = ListFiles(directory);
    
    for i=1:size(files,1),

        i
        
        % Read file
        im = imread(strcat(directory,'/',files(i).name));
     
        % Candidate Generation (pixel) %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
        pixelCandidates = CandidateGenerationPixel_Color(im, pixel_method);
        
        
        % Candidate Generation (window)%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
        % windowCandidates = CandidateGenerationWindow_Example(im, pixelCandidates, window_method); %%'SegmentationCCL' or 'SlidingWindow'  (Needed after Week 3)
        
        % Accumulate pixel performance of the current image %%%%%%%%%%%%%%%%%
        pixelAnnotation = imread(strcat(directory, '/mask/mask.', files(i).name(1:size(files(i).name,2)-3), 'png'))>0;
        [localPixelTP, localPixelFP, localPixelFN, localPixelTN] = PerformanceAccumulationPixel(pixelCandidates, pixelAnnotation);
        pixelTP = pixelTP + localPixelTP;
        pixelFP = pixelFP + localPixelFP;
        pixelFN = pixelFN + localPixelFN;
        pixelTN = pixelTN + localPixelTN;
        
        % Accumulate object performance of the current image %%%%%%%%%%%%%%%%  (Needed after Week 3)
        % windowAnnotations = LoadAnnotations(strcat(directory, '/gt/gt.', files(i).name(1:size(files(i).name,2)-3), 'txt'));
        % [localWindowTP, localWindowFN, localWindowFP] = PerformanceAccumulationWindow(windowCandidates, windowAnnotations);
        % windowTP = windowTP + localWindowTP;
        % windowFN = windowFN + localWindowFN;
        % windowFP = windowFP + localWindowFP;
    end

    % Plot performance evaluation
    [pixelPrecision, pixelAccuracy, pixelSpecificity, pixelSensitivity] = PerformanceEvaluationPixel(pixelTP, pixelFP, pixelFN, pixelTN);
    % [windowPrecision, windowAccuracy] = PerformanceEvaluationWindow(windowTP, windowFN, windowFP); % (Needed after Week 3)
    
    [pixelPrecision, pixelAccuracy, pixelSpecificity, pixelSensitivity]
    % [windowPrecision, windowAccuracy]
    
    %profile report
    %profile off
    toc
end
 



%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% CandidateGeneration
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

function [pixelCandidates] = CandidateGenerationPixel_Color(im, space)

    im=double(im);

    switch space
        case 'normrgb'
            pixelCandidates = im(:,:,1)>100;
            
        otherwise
            error('Incorrect color space defined');
            return
    end
end    
    

function [windowCandidates] = CandidateGenerationWindow_Example(im, pixelCandidates, window_method)
    windowCandidates = [ struct('x',double(12),'y',double(17),'w',double(32),'h',double(32)) ];
end  


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Performance Evaluation
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

function PerformanceEvaluationROC(scores, labels, thresholdRange)
    % PerformanceEvaluationROC
    %  ROC Curve with precision and accuracy
    
    roc = [];
	for t=thresholdRange,
        TP=0;
        FP=0;
        for i=1:size(scores,1),
            if scores(i) > t    % scored positive
                if labels(i)==1 % labeled positive
                    TP=TP+1;
                else            % labeled negative
                    FP=FP+1;
                end
            else                % scored negative
                if labels(i)==1 % labeled positive
                    FN = FN+1;
                else            % labeled negative
                    TN = TN+1;
                end
            end
        end
        
        precision = TP / (TP+FP+FN+TN);
        accuracy = TP / (TP+FN+FP);
        
        roc = [roc ; precision accuracy];
    end

    plot(roc);
end