Puma 2 sequence was taken by connecting a camera to the end of a puma robot arm. The robot arm was then rotated for 120 degrees and 30 image frames were taken each frame being taken approximately after every 4 degrees. The radius of the robot arm was approximately 1.8 feet. The scene was the robotics lab at UMass plastered with posters on the walls and floor. The sequence was taken by Rakesh Kumar and Harpreet Singh Sawhney in Spring 1990 at UMass, Amherst.
Sequence is also described in the paper (Fig. 5) : "Pose Refinement: Application to Model Extension and Sensitivity to camera parameters" by Rakesh Kumar and Allen Hanson. Paper was published in DARPA IU Workshop 1990 (pp. 660-670). Another version was also published in ICCV, Osaka 1990 (pp. 365-369, Fig. 3).
The 30 image frames are in the following directory vis::visimg$root:[puma2]
There are both level 9 (512x512) and level 8 (256x256) images
in that directory. All images are in UMass-LLVS format.
The level 9 images are from fr1.plane to fr30.plane
The level 8 images are from fr1_lev8.plane to fr30_lev8.plane
We did all our experiments on the level 8 images. The output of running
Anandan's optic flow routines are in the directory vis::visimg$root:[puma2.flow].
This was used to do the tracking of image points got by line intersections
(Williams and Hanson, DARPA IU 1987). The *dc.plane files are
the column component of the optic flow for each consecutive pair of image
frames. The *dr.plane files are the row component of the optic
flow for each consecutive pair of image frames.
Camera Specifications:
Imagesizex 256 Imagesizey 242 (for the level 8 images)
Field of view x axis (fovx) 41.67491 degrees
Field of view y axis (fovy) 39.52927 degrees
Focal length in pixels for x axis is given by Sx = (Imagesizex/2) cotan(fovx/2)
Focal length in pixels for y axis is given by Sy = (Imagesizey/2) cotan(fovy/2)
Right handed camera coordinate system.
Optical axis of the camera (pointing out of the image plane) is the z-axis.
Vertical axis (-ve row direction: pointing upward) is y-axis.
Horizontal axis (-ve column direction) is the x-axis.
Center of image plane assumed to be at frame center LLVS coordinates
(row 134.5, col 127.5).
Note the top 14 lines for level 8 images are blank. Therefore the frame
center for rows is got by:
row_cen = (256 - 14)/2 + 14 - 0.5 = 134.5.The frame center for columns
is obtained by:
col_cen = 256/2 - 0.5 = 127.5
Not calibrated for lens distortion. There is some radial lens distortion.
LLVS (row,col) coordinate system has its origin at the top left corner
of the image. Pixel locations for level 8 images range from [(0..255),(0..255)].
To convert from (row,col) coordinates to camera centered (x,y) coordinates.
Use the following equations for level 8 images.
x = 127.5 - col
y = 134.5 - row
Location of the points in the figure for the puma sequence (Fig. 5) shown in paper "Pose Refinement: Application to Model Extension and Sensitivity to camera parameters" by Rakesh Kumar and Allen Hanson. Paper was published in DARPA IU Workshop 1990 (pp. 660-670). Another version was also published in ICCV, Osaka 1990 (pp. 365-369, Fig. 3).
The location of the points are in world coordinates. Right handed world coordinate system.
x-y-z axis correspond to the natural coordinate axis of the room. The origin is at the bottom visible baseboard corner of the room. The y-axis is the vertical (gravity) direction. The z-axis is horizontal direction along the wall coming towards the camera. The x-axis is horizontal direction along the far wall almost perpendicular to the optical axis. We assumed the natural axes of the room were all perpendicular to each other. Measurements done using tape measure.
Location (world coordinates) of the 12 crossed points in Fig. 5.
These points were used to do pose refinement.
| -7.245 | 6.455 |
| -4.22 | 8.09 |
| -3.44 | 7.00 |
| -1.75 | 2.855 |
| -6.94 | -0.47 |
| -7.53 | -0.47 |
| -4.81 | -0.47 |
| -4.82 | -0.47 |
| 0.0 | 5.37 |
| 0.0 | 4.68 |
| 0.0 | 4.12 |
| 0.0 | 8.14 |
Location (world coordinates) of the 20 numbered and circled points in Fig. 5 These points were used for model extension.
| -3.35 | 9.01 | 7.025 |
| -1.495 | 9.01 | 7.035 |
| -3.34 | 9.01 | 3.13 |
| -1.58 | 9.01 | 11.12 |
| -4.31 | 6.45 | 0.0 |
| -6.335 | 8.125 | 0.0 |
| -3.39 | 2.875 | 0.0 |
| -2.22 | 2.45 | 0.0 |
| -6.94 | -0.47 | 16.81 |
| -4.81 | -0.47 | 16.95 |
| -6.44 | -0.47 | 16.95 |
| -6.94 | -0.47 | 17.82 |
| 0.0 | 7.00 | 11.76 |
| 0.0 | 4.69 | 14.91 |
| 0.0 | 4.17 | 11.95 |
| 0.0 | 7.32 | 13.51 |
| -1.30 | 4.15 | 11.28 |
| -2.94 | 4.13 | 11.28 |
| -1.38 | 2.77 | 11.28 |
| -3.02 | 2.75 | 11.28 |
Pose Results using Kumar's R_and_T algorithm are for 30 frames
in order from frame 1 to frame 30.
The results are in order of translation and rotation.
Xc = R(Xw) + T; where Xc is a 3D point
in camera coordinates,Xw is a 3D point in world coordinates,R
is rotation represented as a 4-tuple unit quaternion, and T is translation
(3-tuple).
A quaternion is represented as a 4-tuple: (cos(a/2),sin(a/2)*w) where a
is the angle of rotation and w is the axis of rotation.
Frame 1
0.047770 -3.113052 32.571674 Translation
-0.195041 0.379431 0.901763 0.069379 Rotation
Frame 2
0.301283 -3.265998 32.538456 Translation
-0.198183 0.347196 0.914432 0.063180 Rotation
Frame 3
0.530241 -3.407734 32.523152 Translation
-0.200305 0.314319 0.926175 0.057284 Rotation
Frame 4
0.778154 -3.536952 32.498739 Translation
-0.202630 0.281026 0.936662 0.051278 Rotation
Frame 5
1.045635 -3.655416 32.464440 Translation
-0.205297 0.247138 0.945897 0.045339 Rotation
Frame 6
1.302183 -3.738650 32.464123 Translation
-0.206637 0.213389 0.954071 0.038924 Rotation
Frame 7
1.566383 -3.813139 32.416901 Translation
-0.208329 0.179244 0.960930 0.032936 Rotation
Frame 8
1.841555 -3.867918 32.388667 Translation
-0.209997 0.145036 0.966513 0.026823 Rotation
Frame 9
2.090857 -3.884957 32.402837 Translation
-0.210211 0.111259 0.971099 0.019987 Rotation
Frame 10
2.361540 -3.897319 32.369608 Translation
-0.211278 0.077055 0.974287 0.013734 Rotation
Frame 11
2.646599 -3.891608 32.341944 Translation
-0.211658 0.041715 0.976426 0.007326 Rotation
Frame 12
2.916862 -3.854799 32.347484 Translation
-0.211967 0.008337 0.977241 0.000626 Rotation
Frame 13
3.174283 -3.809674 32.305329 Translation
-0.211611 -0.026997 0.976965 -0.005688 Rotation
Frame 14
3.409921 -3.730540 32.322439 Translation
-0.210904 -0.060345 0.975560 -0.012638 Rotation
Frame 15
3.657849 -3.647953 32.276411 Translation
-0.209941 -0.095189 0.972887 -0.018811 Rotation
Frame 16
3.895837 -3.536705 32.302138 Translation
-0.208900 -0.129340 0.968999 -0.025918 Rotation
Frame 17
4.125824 -3.411148 32.293660 Translation
-0.207518 -0.162930 0.964017 -0.032587 Rotation
Frame 18
4.341265 -3.271096 32.287447 Translation
-0.206132 -0.196149 0.957858 -0.039282 Rotation
Frame 19
4.539352 -3.113609 32.288266 Translation
-0.204132 -0.229277 0.950600 -0.046053 Rotation
Frame 20
4.720604 -2.952910 32.283727 Translation
0.201742 0.262125 -0.942248 0.052528 Rotation
Frame 21
4.876937 -2.779959 32.242796 Translation
-0.198476 -0.294569 0.932928 -0.059012 Rotation
Frame 22
5.042041 -2.592764 32.213589 Translation
-0.195590 -0.326829 0.922321 -0.065209 Rotation
Frame 23
5.194049 -2.400240 32.227964 Translation
-0.192661 -0.358427 0.910650 -0.071610 Rotation
Frame 24
5.322834 -2.192178 32.239625 Translation
-0.189159 -0.389845 0.897864 -0.077970 Rotation
Frame 25
5.453729 -1.960841 32.271508 Translation
-0.185863 -0.420625 0.883943 -0.084697 Rotation
Frame 26
5.557770 -1.715600 32.288481 Translation
-0.182607 -0.451171 0.868771 -0.091300 Rotation
Frame 27
5.634969 -1.481534 32.298767 Translation
-0.177988 -0.480714 0.853051 -0.097660 Rotation
Frame 28
5.703976 -1.224287 32.324813 Translation
-0.173892 -0.510177 0.835844 -0.104141 Rotation
Frame 29
5.747901 -0.982183 32.348424 Translation
-0.169307 -0.538304 0.818174 -0.110254 Rotation
Frame 30
5.772975 -0.726642 32.393893 Translation
-0.164674 -0.566295 0.799112 -0.116675 Rotation