1%% Reset everything 2 3clear all; 4clc; 5close all; 6addpath('helpers'); 7 8%% Configure the benchmark 9 10% noncentral case 11cam_number = 4; 12% let's only get 6 points, and generate new ones in each iteration 13pt_number = 50; 14% noise test, so no outliers 15outlier_fraction = 0.0; 16% repeat 5000 iterations per noise level 17iterations = 5000; 18 19% The algorithms we want to test 20algorithms = { 'gp3p'; 'gpnp'; 'gpnp'; 'abs_nonlin_noncentral'; 'abs_nonlin_noncentral'; 'upnp'; 'upnp' }; 21% This defines the number of points used for every algorithm 22indices = { [1, 2, 3]; [1:1:10]; [1:1:50]; [1:1:10]; [1:1:50]; [1:1:10]; [1:1:50] }; 23% The name of the algorithms on the plots 24names = { 'GP3P'; 'GPnP (10pts)'; 'GPnP (50pts)'; 'nonlin. opt. (10pts)'; 'nonlin. opt. (50pts)'; 'UPnP (10pts)'; 'UPnP (50pts)' }; 25 26% The maximum noise to analyze 27max_noise = 5.0; 28% The step in between different noise levels 29noise_step = 0.1; 30 31%% Run the benchmark 32 33%prepare the overall result arrays 34number_noise_levels = max_noise / noise_step + 1; 35num_algorithms = size(algorithms,1); 36mean_position_errors = zeros(num_algorithms,number_noise_levels); 37mean_rotation_errors = zeros(num_algorithms,number_noise_levels); 38median_position_errors = zeros(num_algorithms,number_noise_levels); 39median_rotation_errors = zeros(num_algorithms,number_noise_levels); 40noise_levels = zeros(1,number_noise_levels); 41 42%Run the experiment 43for n=1:number_noise_levels 44 45 noise = (n - 1) * noise_step; 46 noise_levels(1,n) = noise; 47 display(['Analyzing noise level: ' num2str(noise)]) 48 49 position_errors = zeros(num_algorithms,iterations); 50 rotation_errors = zeros(num_algorithms,iterations); 51 52 counter = 0; 53 54 validIterations = 0; 55 56 for i=1:iterations 57 58 % generate experiment 59 [points,v,t,R] = create2D3DExperiment(pt_number,cam_number,noise,outlier_fraction); 60 [t_perturbed,R_perturbed] = perturb(t,R,0.01); 61 T_perturbed = [R_perturbed,t_perturbed]; 62 T_gt = [R,t]; 63 64 % run all algorithms 65 allValid = 1; 66 67 for a=1:num_algorithms 68 T = opengv(algorithms{a},indices{a},points,v,T_perturbed); 69 [position_error, rotation_error] = evaluateTransformationError( T_gt, T ); 70 71 if( position_error > 100 ) 72 allValid = 0; 73 break; 74 else 75 position_errors(a,validIterations+1) = position_error; 76 rotation_errors(a,validIterations+1) = rotation_error; 77 end 78 end 79 80 if allValid == 1 81 validIterations = validIterations +1; 82 end 83 84 counter = counter + 1; 85 if counter == 100 86 counter = 0; 87 display(['Iteration ' num2str(i) ' of ' num2str(iterations) '(noise level ' num2str(noise) ')']); 88 end 89 end 90 91 %Now compute the mean and median value of the error for each algorithm 92 for a=1:num_algorithms 93 mean_position_errors(a,n) = mean(position_errors(a,1:validIterations)); 94 median_position_errors(a,n) = median(position_errors(a,1:validIterations)); 95 mean_rotation_errors(a,n) = mean(rotation_errors(a,1:validIterations)); 96 median_rotation_errors(a,n) = median(rotation_errors(a,1:validIterations)); 97 end 98 99end 100 101%% Plot the results 102 103figure(1) 104plot(noise_levels',mean_rotation_errors','LineWidth',2) 105legend(names,'Location','NorthWest') 106xlabel('noise level [pix]') 107ylabel('mean rot. error [rad]') 108grid on 109 110figure(2) 111plot(noise_levels',median_rotation_errors','LineWidth',2) 112legend(names,'Location','NorthWest') 113xlabel('noise level [pix]') 114ylabel('median rot. error [rad]') 115grid on 116 117figure(3) 118plot(noise_levels',mean_position_errors','LineWidth',2) 119legend(names,'Location','NorthWest') 120xlabel('noise level [pix]') 121ylabel('mean pos. error [m]') 122grid on 123 124figure(4) 125plot(noise_levels',median_position_errors','LineWidth',2) 126legend(names,'Location','NorthWest') 127xlabel('noise level [pix]') 128ylabel('median pos. error [m]') 129grid on