GENERAL INFORMATION October 5 1992 : Greg Lee and George Stockman ------------------- This file is available as /pub/range_images/README by anonymous FTP to pixel.cps.msu.edu at the PRIP Lab of Michigan State University. This directory contains numerous fused range and intensity images produced at the Michigan State University Pattern Recognition and Image Processing (PRIP) laboratory. You are free to copy, distribute and use these images to test your algorithms. However, if the images are to appear in a published article, please acknowledge the MSU PRIP Lab as the source of the images. Our scenes are very similar to those formerly scanned and made available to the public by Pat Flynn -- the difference is that in this new data there are real intensity values available for the surface points. A technical report giving more details about this data set will be forthcoming : a Latex file containing the text should appear in this directory sometime in the Fall 1991. A paper copy will be available to those who request it and will, in addition to the text, contain our best efforts at reproducing the intensity and range images using grey scale. The fused range/intensity images have been acquired via use of the White Scanner from Technical Arts as explained below. For convenience in its use, the data is presented as if it were formed from the scene using parallel projection along the optical axis of the camera. However, a perspective transformation was used in order to map 3D world surface points onto the intensity array produced by the camera, so that intensities could be assigned to 3D surface points. The resulting data is definitely not 3D, but rather 2 1/2 D, and is very much viewpoint dependent. Because the White Scanner is a triangulating sensor, there are voids in the data due to shadowing from both the camera and laser sheet. Shadowing from the white light sources has been minimized by using two or three light sources when taking the intensity images. Care has been taken to assure registration accuracy to within one pixel in the final fused images. The fusion process is brifely outlined here. Complete details will be available in the near future when the technical report is produced. 1) Obtain range image using Technical Arts White Scanner. Data in this range image (Xw, Yw, Zw) is in the World coordinate frame. 2) Obtain intensity image (Ic) using Innovision's color imaging system through White Scanner's camera (with white lights on and laser off). 3) Find > 6 corresponding points in range and intensity images and compute the calibration matrix for transforming a point in 3D to the camera image plane. 4) Compute the transformation matrix for aligning the axes of world and camera coordinate systems. 5) Parallel project the (Xw,Yw,Zw) into the fused image plane. (This is a synthetic image plane that is coplaner with the CCD array of the camera.) Each element in the fused image plane has 3D coordinates (Xc,Yc,Zc), from which depth can easily be computed, and the corresponding intensity value. The intensity value at (Xc,Yc,Zc) is taken as the intensity value at the pixel in the original intensity image which results from projecting (Xw,Yw,Zw) using the calibration matrix computed in step 3). 6) Only those pixels in the fused image having both range and intensity values are valid. A binary flag image is used to indicate the validity of the range/intensity values at each element in the image. Thus the final fused image has 5 components (Flag, X, Y, Z, Intensity). Because we have forward transformed the data, there are often small holes or lines in the fused data. These are due to quantization and not shadowing. Most such holes may be easily removed using a median filter (or, by doing back transformation and interpolation). We thought that it might be better for scientific research to view the data as spatial samples which may not be dense in the array, and, thus we have not filled the holes at this time. FILE FORMAT ----------- The file format follows that of Pat Flynn's range images in Vision List Archive. All images are in compressed ASCII text format. The first two lines of each image gives the number of rows and columns of the image. The third line is a comment line. This is followed by five images (flag,x,y,z,inten). The first image is the 'flag' (binary) image, where a pixel value of 1 means the corresponding (x,y,z,inten) values are valid; else, the (x,y,z,inten) components should be ignored. Following the 'flag' image are the images of X-coordinates, Y-coordinates, and Z-coordinates in sequence. All are floating-point images. The last image, 'inten', captures the registed intensity value at each pixel. The intensity values are scaled and represented by integers between 0 and 255. IMAGE SURFACE TYPES ------------------- file name surface types comments ------------ ------------------------------ -------- adaptor1 planar,cylindrical adaptor2 planar,cylindrical agpart1 planar,cylindrical agpart2 planar,cylindrical SPIE.1 block1 planar block1+column1 planar,cylindrical SPIE.2 block2 planar,cylindrical SPIE.2 bulb1 spherical,cylindrical SPIE.1 cap1 planar,cylindrical column1 planar,cylindrical column2 planar,cylindrical cone+cylinder cylindrical,conical SPIE.2 cone+cylinder+sphere spherical,cylindrical,conical SPIE.2 cone+sphere spherical,conical cone1 conical SPIE.1 cone2 planar,conical cup1 cylindrical,(????) cup2 planar,cylindrical,(????) cylinder+sphere planar,spherical,cylindrical cylinder1 cylindrical cylinder2 planar,cylindrical SPIE.1 funnel1 conical funnel2 conical gb1 planar gb2 planar halfsphere planar,spherical hump+agpart planar,cylindrical,(occlusion) SPIE.2 hump1 planar,cylindrical jig1 planar SPIE.1 jig2 planar pipe1 cylindrical pipe2 cylindrical propane planar,cylindrical roof1 planar roof2 planar sphere spherical y_big1 cylindrical y_small1 cylindrical PROBLEMS and COMMENTS --------------------- Please send E-mail to manager@pixel.cps.msu.edu