3D Coordinates Line of Fit

73 ビュー (過去 30 日間)
David Robie
David Robie 2018 年 6 月 20 日
コメント済み: Wes Anderson 2019 年 12 月 3 日
Hello, I have an Nx3 matrix which represents sets of coordinates in 3D space. Is there a way to calculate a line of best fit (or any type of regression) to generate an equation for approximating expected data points? Thank you!

採用された回答

Anton Semechko
Anton Semechko 2018 年 6 月 20 日
Hey, David. Here is a demo you may find useful:
function best_fit_3D_line_demo
% Generate sample dataset
% -------------------------------------------------------------------------
% Standard deviation of noise
s=1;
% Random direction in space
r=randn(3,1);
r=r/(norm(r)+eps);
% Random data points along r
N=1E3; % number of point samples
t=(10*s)*(2*rand(N,1)-1);
Xo=bsxfun(@times,t,r');
% Add (isotropic) Gaussian noise to Xo
X=Xo+s*randn(N,3);
% Offset X (relative to the origin) by a random amount
X=bsxfun(@plus,X,5*s*randn(1,3));
% Find line of best fit (in least-squares sense) through X
% -------------------------------------------------------------------------
X_ave=mean(X,1); % mean; line of best fit will pass through this point
dX=bsxfun(@minus,X,X_ave); % residuals
C=(dX'*dX)/(N-1); % variance-covariance matrix of X
[R,D]=svd(C,0); % singular value decomposition of C; C=R*D*R'
% NOTES:
% 1) Direction of best fit line corresponds to R(:,1)
% 2) R(:,1) is the direction of maximum variances of dX
% 3) D(1,1) is the variance of dX after projection on R(:,1)
% 4) Parametric equation of best fit line: L(t)=X_ave+t*R(:,1)', where t is a real number
% 5) Total variance of X = trace(D)
% Coefficient of determineation; R^2 = (explained variance)/(total variance)
D=diag(D);
R2=D(1)/sum(D);
% Visualize X and line of best fit
% -------------------------------------------------------------------------
% End-points of a best-fit line (segment); used for visualization only
x=dX*R(:,1); % project residuals on R(:,1)
x_min=min(x);
x_max=max(x);
dx=x_max-x_min;
Xa=(x_min-0.05*dx)*R(:,1)' + X_ave;
Xb=(x_max+0.05*dx)*R(:,1)' + X_ave;
X_end=[Xa;Xb];
figure('color','w')
axis equal
hold on
plot3(X_end(:,1),X_end(:,2),X_end(:,3),'-r','LineWidth',3) % best fit line
plot3(X(:,1),X(:,2),X(:,3),'.k','MarkerSize',13) % simulated noisy data
set(get(gca,'Title'),'String',sprintf('R^2 = %.3f',R2),'FontSize',25,'FontWeight','normal')
xlabel('X','FontSize',20,'Color','k')
ylabel('Y','FontSize',20,'Color','k')
zlabel('Z','FontSize',20,'Color','k')
view([20 20])
drawnow
  9 件のコメント
Wes Anderson
Wes Anderson 2019 年 12 月 3 日
Hi,
I have the same issue, but I'd like my regression to go from the axis to those data points, and to get an equation out of it...any idea how to do that?
Thank you
Wes Anderson
Wes Anderson 2019 年 12 月 3 日
*from the origin, not the axis. Sorry.

サインインしてコメントする。

その他の回答 (0 件)

カテゴリ

Help Center および File ExchangeLinear Algebra についてさらに検索

製品


リリース

R2018a

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by