Data Set Overview:This data set represents the primary results of three dimensional(3-D) modeling of the Mars Pathfinder landing site using data from theImager for Mars Pathfinder camera. The camera system is described bySmith et al. [SMITHETAL1997A, SMITHETAL1997B]. It consisted of astereo imager pair located on a pan and tilt platform. Each imager ofthe pair was equipped with a filter wheel so that the camera set couldimage the landscape in 15 narrow-band filters.This data set consists of a set of tables. Each table contains 3-Dobject position information in the form of a Cartesian (x,y,z)coordinate in units of meters corresponding to each pixel in an IMPEDR stereo pair acquired in the 670 nm filter. The coordinates arededuced using an automated machine vision algorithm that correlatesfeatures between the left and right images of stereo pairs todetermine their disparity (difference in image position between theleft and right eye) then computes their 3-D object position takinginto account the camera pointing and stereo optics. The computeralgorithm is described by Stoker et al. [STOKERETAL1999] andsummarized below.Stereo model products (and corresponding tables) have been producedfor two IMP Pathfinder data sets acquired in stereo in the 670 nmfilter. The IMP data sets are described by Gaddis etal. [GADDISETAL1999]. The stereo data sets that were analyzed arecalled the Monster Pan and the Super Pan. The Monster Pan was acomplete stereo panorama of the Pathfinder landing site acquired earlyin the mission (sols 3-6). The monster pan images in the 670 nmfilter were compressed using lossy JPEG compression (6:1 compressionfactor) and the image to image overlap in the panoramic product wasrelatively low. The Super Pan was designed to produce a full panoramaof the landing site with low compression ratio in all 15 narrow-bandfilters and the 670 nm stereo filter set was losslessly compressedusing Rice compression. It was designed with increased frame-to-frameoverlap relative to the Monster Pan to assist with automated matchingbetween images and insure gap-free stereo coverage. The Super Panrepresented a large data volume and was acquired over an 8 week periodfrom sols 13 to 80. It was 83% complete when the mission ended.While incomplete, the 3-D reconstructions from the Super Pan imagesare somewhat better than for the Monster Pan due to the increasedimage overlap and lower image compression.Parameters:Each table entry consists of a Cartesian coordinate corresponding tothe object position computed for each pixel of the left member of astereo pair for which a model solution was obtained. The origin ofthe coordinate system for the values provided in these tables is atthe intersection of the camera elevation and azimuth axes. The X axisis aligned with north so that +X values are north of the origin, the Yaxis is aligned with west so that +Y values are west of the origin,and the +Z direction is up. Other parameters which enter into thestereo reconstruction are described below along with a description ofthe Stereo Pipeline algorithm.Processing:The computer program which produces the 3-D reconstruction is calledthe Ames Stereo Pipeline [STOKERETAL1999]. The input to thestereo matching algorithm consists of raw EDR images from an IMPstereo pair. Results are better if images are used that have not beenflat field corrected or photometrically calibrated because theseprocesses resample the pixel information. The first stage in theStereo Pipeline algorithm is called the 'preprocessing' stage andinvolves preparing the input stereo pair to improve the correlation inthe later stages. First, a linear stretch is applied to normalize theimage intensity between the left and right members of the stereo pair.This is needed because the correlation algorithm works by matching theintensity values between the image pairs. Then, a uni-directionalSobel edge enhancement technique [BAXES1994] is applied. Nextcalculations are performed to correct for translational, rotational,and pixel-scale differences between the left and right eyes.The next stage of processing in the Stereo Pipeline is to correlatethe features in the images between the left and right cameras. Theresult of this stage is a disparity calculation for each pixel in theimage pair. A texture-based sum-of-absolute-difference (SOAD)correlation algorithm is used and the consistency of each match isvalidated by doing both a correlation and cross-correlation. Thisalmost eliminates matches between wrong local figures. A smallsubframe of the image surrounding a considered pixel, called thekernel, is selected from one member of the stereo pair. The kernel isslid over the other image of the pair by a step of one pixel at atime, a subtraction is performed, and the elements of the resultingmatrix are summed. This procedure is used to find the position of themost similar portion of the test image with the kernel. Threecorrelation passes, using different sized kernels, are used to improveboth computational speed and accuracy. The same correlation algorithm,with different parameters, is used for all three passes. The firstpass of the correlator is used to bound the disparity range of theimage. It uses a small kernel and searches across the complete rangeof possible disparity values. For this first pass, a relatively lowrate of correlations are found, but these are used to limit the searchspace of the disparity for the next pass. The second correlation passuses a larger kernel which results in a high percentage of pixelsbeing matched. In the final (third) pass, the disparity search isconstrained to the neighborhood of the disparity calculated in theprevious pass. Ideally, a small kernel size is preferred for this passbecause the disparity value assigned to the pixel is the average overthe kernel. Kernel size for the second and third pass are userselectable. For the models published here, kernel size wasinteractively varied to minimize the amount of pixel-to-pixel variancein computed 3-D position. High variance results from errors in theestimate of disparity. Small errors in the estimate of disparity canlead to large errors in the estimate of position along the camera'sline of site. The Kernel size for the second and third correlationpass is a user defined quantity of n columns by m rows. Values usedfor this data set were 14x14 pixels (second pass) and 27x27 pixels(third pass). The correlation stage is followed by a filtering stagethat removes 'outliers'-- disparity values much different than thosein the nearby area. Next, gaps in the disparity map are filled. Gapsare places which had no match, inconsistent cross-correlations, oroutlier disparities. Some gaps are the result of real-worlddiscontinuities in surface shape, such as the occluding boundaries ofrocks in the terrain. In order to retain these boundaries in the map,gaps occurring at large discontinuities are filled with the minimumdisparity value (corresponding to the point furthest from the camera)in the gap neighborhood. Gaps in regions with small disparity varianceare more likely due to a smooth, texture free surface. These gaps canbe filled by interpolation or set to zero. In the models publishedhere they are set to zero to avoid confusing them with values computedby the algorithm. The next processing stage derives 3-D positionpoints from disparity values. Each pixel is projected along a vectordefined by the (line, sample) coordinate of the pixel and the nodalpoint of the camera to a distance consistent with its disparity. Thisintersection point is the object coordinate. Then, using the camerapan and tilt angle, the object coordinates are rotated to the landercoordinate system. This computation is repeated for each pixel of thestereo pair to get a set of object points. These object points, intabular form, are the data set provided.Ancillary Data:The data are referenced to raw IMP EDR images. These will be requiredfor interpretation of the 3-D model data.

Data Coverage and Quality:As discussed above, the 3-D position information is deduced matchingbrightness patterns in the left and right eyes of the stereo pair.When no match is found, or inconsistent matches found in thecorrelation and cross correlation, no disparity is calculated and avalue of zero is assigned to the Cartesian coordinate (X:Y:Z:0) inthe table. Thus, zero values in the table indicate that the stereomatching algorithm did not yield a good solution at that location.Confidence Level and Limitations:For the Mars Pathfinder IMP camera data sets, the error in the 3-Dposition of an object point in the model comes from the followingsources:1) The uncertainty in the azimuth and elevation of the camera leads touncertainty in the 3-D model position. According to the IMPcalibration report [CROWEETAL1996] the pointing error acts in a planeperpendicular to the camera optical axis. This error is a linearfunction of the camera-point distance and is within +/- 2.7% inazimuth and +/- 1.2% in elevation of the absolute position of thepoint (assuming a pan error of +/- 1.5 degrees and a tilt error of +/-0.65 degrees. These are worst case values due to backlash in thecamera motors.Of the uncertainty sources, this is the largest, but the camerapointing uncertainty affects all points from one stereo pair equallyas a solid body. This source of error can be minimized by determiningactual camera pointing after the fact by using tiepoints betweenstereo pairs. The United States Geological Survey Astrogeology Branch,under the direction of R. Kirk, undertook a project to provideimproved camera pointing information using a control network for thesite and bundle adjustment. This procedure is described by Kirk etal. [KIRKETAL2001]. The values they determined were substituted forsurface based instrument azimuth and elevation for the instrumenttelemetry values provided in the original EDR headers. Inspection ofthe results showed that using these values produced a noticeableimprovement in how well models from adjacent images fit together. Theyalso computed values for left toe-in (-13.732 radians), right toe-in(24.63 radians) and boresight angles (1.116 radians) that aredifferent from those published by the IMP camera team [CROWEETAL1996].We also used these values in our computations.2) Uncertainty in the computed camera-point distance results from thedisparity computation method. For any pixel, the computed disparityrepresents an average over the Kernel for the final correlationpass. Smaller Kernel sizes lead to a high percentage of falsecorrelations. Thus the models appear more noisy. Even though adisparity point is assigned to each pixel, the real resolution of themodel is a function of the Kernel size in the final pass.3) Image resolution limits stereo matching precision. Subpixeldisparity is not computed by the algorithm.4) The stereo images of the Monster Pan were compressed using lossyJPEG compression. High correlation rates are achieved even with thecompressed data of the Monster Pan but the results are clearly noisier(defined as pixel to pixel variance in 3-D position computed by thealgorithm) than for the losslessly compressed Super Pan. As discussedabove, this variance is due to errors in the estimated disparity.Compression artifacts result in a higher percentage of false matches.