Robot Vision Systems Camera Model Calibration 1 1 “perspectograph” Alberti’s Grid 1 2 Pinhole Camera scene image plane 1 iris, optical center 3 Pinhole Camera “image coordinates” 1 4 Pinhole Camera “Alberti’s Grid” 1 5 Pinhole Camera • Classical pin-hole r’ 1 x f z 6 Classical Pinhole Camera • Similar Triangles c’ 1 x f z 7 Classical Pinhole Camera • Similar Triangles c’ Image coordinates Projective coordinates 1 x f Camera matrix (projection) z Point expressed in the camera frame 8 Classical Pinhole Camera • Similar Triangles Convert pixels to mm: 1 c’ x f z world 9 Camera Calibration • Used to determine the elements in the camera matrix • Use pairs of known world points and their corresponding image points e.g. use calibration grid 1 10 Camera Model – Perspective Projection • Need to determine the parameters! 1 11 Camera Calibration • camera matrix 1 12 Camera Calibration • Projective Equivalence Two equations in 12 unknowns 1 13 Camera Calibration • Have 6 point pairs (c,r) and (x,y,z) – Correspondence known – World coordinates known accurately 1 14 Camera Calibration 1 • Method 1: assume a34 = 0 and solve for aij • Solve using pseudo-inverse 15 Camera Calibration 1 • Method 2: make no assumption about a34 and solve for aij • The vector a = [a11, …, a34]T is in the nullspace of the design matrix 16 Camera Calibration • Ba = 0, • Use SVD, then a is the column of V corresponding to the null singular value of B – One property of the SVD is that the columns of V corresponding to the zero singular value span the null space of B 1 17 Camera Calibration • Given the camera matrix solved w.r.t. the robot base frame 1 18 Using the Camera Matrix • Projection ( 1 is 3 x 4)– cannot invert! 19 Measurements from Images • Must have relationship between the image “pixels” and the world • 2D imaging – the image plane and the “world” plane are in 1-1 correspondence 1 20 Using the Camera Matrix in 2D • If all world points are on a plane • Then z is a linear function of x & y 1 21 Using the Camera Matrix in 2D • Now the projection equations • Can be written 1 22 Using the Camera Matrix in 2D • Now the projection equations • Can be written 1 23 Using the Camera Matrix in 2D • Now the projection equations • Can be written 1 2 equations, 9 unknowns 24 Using the Camera Matrix in 2D • Four known points, i=1,..4 8 equations, 9 unknowns up to a scale (8 unknowns) • Linear equations -- can be solved 1 25 Using the Camera Matrix in 2D 1 26 Using the Camera Matrix in 2D • Now the projection equations are simpler … and we can “invert” (map image points back to the world plane) 1 27 Measurements from Images 1 28 Robot to Plane • Homogeneous transformation from base (or end-effector) frame to the work-plane of the imaging system – Use “known” corresponding points to solve for the elements of T (6 unknowns) 1 29