class="markdown_views prism-kimbie-light">

slam国内外数据库

（牛逼博主连接：https://www.cnblogs.com/yhlx125/p/5609998.html） http://www.ifp.uni-stuttgart.de/ISPRS-EuroSDR/ImageMatching/index.html http://www.robots.ox.ac.uk/NewCollegeData/index.php?n=Main.Downloads#GettingStarted http://asrl.utias.utoronto.ca/datasets/devon-island-rover-navigation/ http://kos.informatik.uni-osnabrueck.de/3Dscans/

2D激光扫描匹配方法

1.Beam Model

Beam Model我将它叫做测量光束模型。个人理解，它是一种完全的物理模型，只针对激光发出的测量光束建模。将一次测量误差分解为四个误差。 phhitphhit，测量本身产生的误差，符合高斯分布。 phxxphxx，由于存在运动物体产生的误差。 …

2.Likehood field

似然场模型，和测量光束模型相比，考虑了地图的因素。不再是对激光的扫描线物理建模，而是考虑测量到的物体的因素。 似然比模型本身是一个传感器观测模型，之所以可以实现扫描匹配，是通过划分栅格，步进的方式求的最大的Score，将此作为最佳的位姿。 for k=1:size(zt,1) if zt(k,2)>0 d = -grid_dim/2; else d = grid_dim/2; end phi = pi_to_pi(zt(k,2) + x(3)); if zt(k,1) ~= Z_max ppx = [x(1),x(1) + zt(k,1)*cos(phi) + d]; ppy = [x(2),x(2) + zt(k,1)*sin(phi) + d]; end_points = [end_points;ppx(2),ppy(2)]; wm = likelihood_field_range_finder_model(X(j,:)',xsensor,... zt(k,:)',nearest_wall, grid_dim, std_hit,Z_weights,Z_max); W(j) = W(j) * wm; else dist = Z_max + std_hit*randn(1); ppx = [x(1),x(1) + dist*cos(phi) + d]; ppy = [x(2),x(2) + dist*sin(phi) + d]; missed_points = [missed_points;ppx(2),ppy(2)]; end set(handle_sensor_ray(k),'XData', ppx, 'YData', ppy) end function q = likelihood_field_range_finder_model(X,x_sensor,zt,N,dim,std_hit,Zw,z_max) % retorna probabilidad de medida range finder :) % X col, zt col, xsen col [n,m] = size(N); % Robot global position and orientation theta = X(3); % Beam global angle theta_sen = zt(2); phi = pi_to_pi(theta + theta_sen); %Tranf matrix in case sensor has relative position respecto to robot's CG rotS = [cos(theta),-sin(theta);sin(theta),cos(theta)]; % Prob. distros parameters sigmaR = std_hit; zhit = Zw(1); zrand = Zw(2); zmax = Zw(3); % Actual algo q = 1; if zt(1) ~= z_max % get global pos of end point of measument xz = X(1:2) + rotS*x_sensor + zt(1)*[cos(phi); sin(phi)]; xi = floor(xz(1)/dim) + 1; yi = floor(xz(2)/dim) + 1; % if end point doesn't lay inside map: unknown if xi<1 || xi>n || yi<1 || yi>m q = 1.0/z_max; % all measurements equally likely, uniform in range [0-zmax] return end dist2 = N(xi,yi); gd = gauss_1D(0,sigmaR,dist2); q = zhit*gd + zrand/zmax; end end

3.Correlation based sensor models相关分析模型

XX提出了一种用相关函数表达马尔科夫过程的扫描匹配方法。 互相关方法Cross-Correlation，另外相关分析在进行匹配时也可以应用，比如对角度直方图进行互相关分析，计算变换矩阵。 参考文献：A Map Based On Laser scans without geometric interpretation circular Cross-Correlation的Matlab实现 % Computes the circular cross-correlation between two sequences % % a,b the two sequences % normalize if true, normalize in [0,1] % function c = circularCrossCorrelation(a,b,normalize) for k=1:length(a) c(k)=a*b'; b=[b(end),b(1:end-1)]; % circular shift end if normalize minimum = min(c); maximum = max(c); c = (c - minimum) / (maximum-minimum); end 4.MCL 蒙特卡洛方法 5.AngleHistogram 角度直方图 6.ICP/PLICP/MBICP/IDL 属于ICP系列，经典ICP方法，点到线距离ICP， 7.NDT 正态分布变换 8.pIC 结合概率的方法 9.线特征 目前应用线段进行匹配的试验始终不理想：因为线对应容易产生错误，而且累积误差似乎也很明显！

2D激光线特征提取

Nguyen, V., et al. (2007).“A comparison of line extraction algorithms using 2D range data for indoor mobile robotics.” Autonomous Robots 23(2): 97-111. 论文提出了6中从二维激光扫描数据中提取线段的方法 1.分割合并算法

2.回归方法 先聚类，再回归 3.累积、区域生长算法 感觉对噪声数据真的没办法了，窝成一团的点，提取的线十分破碎而且乱… function [ lineSegCoord ] = extractLineSegment( model,normals,intervalPts,normalDelta,dThreshold) %EXTRACTLINESEGMENT Summary of this function goes here % Detailed explanation goes here if (nargin == 0) || isempty(model) lineSegCoord = []; return; end; ns = createns(model','NSMethod','kdtree') pts=size(model,2); if (nargin == 3) normalDelta=0.9; dThreshold=0.5; end if isempty(normals) normals=zeros(2,pts); for nor=1:pts [idx, dist] = knnsearch(ns,model(:,nor)','k',2); data=model(:,idx); men=mean(data,2); rep= repmat(men,1,size(data,2)); data = data - rep; % Compute the MxM covariance matrix A A = cov(data'); % Compute the eigenvector of A [V, LAMBDA] = eig(A); % Find the eigenvector corresponding to the minimum eigenvalue in A % This should always be the first column, but check just in case [~,idx] = min(diag(LAMBDA)); % Normalize V = V(:,idx)./norm(V(:,idx)); %定向 normals(:,nor)=V; end end lineSeg=[1;1]; newLineIdx=1; for j=2:pts-1 current=model(:,j); pre=model(:,j-1); next=model(:,j+1); curNormal=normals(:,j); preNormal=normals(:,j-1); nextNormal=normals(:,j+1); [d,vPt]=Dist2D_Point_to_Line(current,pre,next); dis=norm(current-pre); delta=dot(curNormal,preNormal)/(norm(curNormal)*norm(preNormal)); if(delta>normalDelta&& d intervalPts) try pts= model(:,(from:to)); coef1 = polyfit(pts(1,:),pts(2,:),1); k2 = coef1(1); b2 = coef1(2); coef2 = robustfit(pts(1,:),pts(2,:),'welsch'); k2 = coef2(2); b2 = coef2(1); ML = true; catch ML = false; end; [D,fPb]= Dist2D_Point_to_Line(model(:,from),[0 b2]',[1 k2+b2]'); [D,tPb]= Dist2D_Point_to_Line(model(:,to),[0 b2]',[1 k2+b2]'); interval=abs(model(1,from) -model(1,to)); if(interval>0.05) x = linspace(fPb(1) ,tPb(1), 5); if ML y_ML = k2*x +b2; lineSegCoord=[lineSegCoord [fPb(1) fPb(2) tPb(1) tPb(2)]']; plot(x, y_ML, 'b-', 'LineWidth', 1); end; else y = linspace(fPb(2) ,tPb(2), 5); if ML x_ML =(y-b2)/k2; lineSegCoord=[lineSegCoord [fPb(1) fPb(2) tPb(1) tPb(2)]']; plot(x_ML, y, 'b-', 'LineWidth', 1); end; end; % try % tmpPts= model(:,(from:to)); % Theta_ML = estimate_line_ML(tmpPts,[], sigma, 0); % ML = true; % catch % % probably the optimization toolbox is not installed % ML = false; % end; % interval=abs(model(1,from) -model(1,to)); % if(interval>10) % x = linspace(model(1,from) ,model(1,to), 5); % if ML % y_ML = -Theta_ML(1)/Theta_ML(2)*x - Theta_ML(3)/Theta_ML(2); % lineSegCoord=[lineSegCoord [x(1) y_ML(1) x(5) y_ML(5)]']; % plot(x, y_ML, 'b-', 'LineWidth', 1); % end; % else % y = linspace(model(2,from) ,model(2,to), 5); % if ML % x_ML = -Theta_ML(2)/Theta_ML(1)*y - Theta_ML(3)/Theta_ML(1); % lineSegCoord=[lineSegCoord [x_ML(1) y(1) x_ML(5) y(5)]']; % plot(x_ML, y, 'b-', 'LineWidth', 1); % end; % end; end end end 4.Ransac方法 5.霍夫变换方法 6.EM方法