Instrument Overview
  ===================
    The rover, named 'Sojourner', was essential to the achievement of
    the major scientific goals of the Pathfinder mission because it
    carried the Alpha Proton X-ray Spectrometer, or APXS, that
    analyzed rocks and other surface materials.  Aside from
    transporting the APXS, the rover also contributed scientifically
    by carrying three cameras that yielded a large quantity of image
    data.  The discussion of the rover 'instrument' per se in this
    section will focus on the salient characteristics of the three
    cameras, and not of the APXS.  For a complete discussion of the
    latter, see the instrument description for the APXS.
 
    Two front monochromatic camera/laser systems provided stereo
    viewing and ranging, while the rear mounted camera was
    discriminant in the red, green, and IR spectral bandwidths to
    produce color images.  Whereas the Imager for Mars Pathfinder
    (IMP) camera viewed the Martian scenes from a fixed platform at
    spatial resolutions that diminished with distance, the mobile
    rover was able to drive up to landforms, including those hidden
    from the IMP, to image them at higher resolution.
 
    The rover cameras were built by the Microrover Flight Experiment
    Team at the Jet Propulsion Laboratory.
 
 
  Scientific Objectives
  =====================
    The rover's Primary Mission was planned for seven sols (Martian
    days).  The rover was designed to interact with slightly more
    than 300 square meters of Martian terrain, though initial
    operations were conducted within a limit of 10 meters or less of
    the lander.
 
    Primary objectives for the rover were 1) exit the lander as early
    as practical on the basis of lander stereoscopic images, 2) send
    initial vehicle performance and technology experiment data to the
    lander, 3) move a few meters and repeat objective #2, 4) acquire
    and transmit images showing the condition of the lander, 5)
    acquire images at the end of daily traverses for navigation
    purposes, or encounter, acquire, and transmit an image of a rock
    or soil patch for subsequent APXS analysis, 6) deploy APXS on the
    imaged rock or soil patch, if possible, for 1 to 10 hours
    duration for chemical analysis, 7) query the APXS for final data
    and transmit interim and final data to the lander, and 8)
    traverse diverse terrains and repeat objective #2.
 
    An Extended Mission of 30 sols was planned and more than
    realized, as the governing spacecraft, rover and site-related
    factors permitted rover activities to continue for 81 sols.
    Extended Mission objectives were similar to the Primary
    objectives, with technology experiments amended to explore the
    diverse terrains further away from the lander.
 
    As during the Primary Mission, Extended operations of the rover
    cameras were to yield a high resolution image dataset critical
    for navigation purposes and geologic analysis of structures in
    rocks and soil-like materials.  The rover image data were to
    contribute to the technology experiments by affording high
    resolution stereoscopic and multispectral examination of 1)
    features targeted for APXS chemical analysis, 2) slumped or
    eroded surface material that had been excavated by the churning
    rover wheels, 3) aeolian effects on the exposed, upturned
    material, and 4) overturned rocks.  This would aid in achieving
    the scientific objective of better understanding surface material
    properties such as grain size, bulk density, friction angle,
    cohesion, and compressibility, which could then be put in the
    larger context of geologic features seen in the lander IMP
    images.
 
    Of the ten Technology Experiments, rover camera images were
    acquired to support the following six: a) Terrain
    Characterization, b) Basic Soil Mechanics, c) Wheel Abrasion, d)
    Thermal Characterization, e) Dead Reckoning and Path
    Reconstruction, and f) Vision Sensor Performance.  For a
    description of these experiments, refer to [MATIJEVICETAL1997A].
 
 
  Subsystems
  ==========
    There were three imaging subsystems aboard the rover.  Two
    broad-band monochrome camera/laser systems were located on the
    forward part of the rover, and a third camera was located on the
    aft part near the APXS.  The two front cameras, used in
    conjunction with laser stripe projectors for hazard detection,
    provided images for stereoscopy and measurements.  The aft
    camera, which was rotated 90 degrees, provided spectral
    information while imaging the APXS target area, rover tracks, and
    terrain.
 
    The 'cameras' were CCDs clocked out by the rover Central
    Processing Unit (CPU).  The camera systems were capable of
    auto-exposure, block truncation coding (BTC) data compression,
    bad pixel/column handling, and image data packetizing.
 
    The major components of the camera systems are described in
    detail below.
 
 
  Detectors
  =========
    Front Camera Specifications:
 
    CCD array size        : 484 (vertical) x 768 (horizontal) pixels
    Pixel size            : 13.6 um (vertical) x 11.6 um (horizontal)
    Full well             : 60,000 electrons
                            (manufacturer claims 80,000)
    Output sensitivity    : 10 uV/electron
    Dynamic range         : > 60 dB
    Dark Noise            : 40 electrons rms
    Dark Current          : < 0.5 nA/cm**2
    Blooming suppression  : > 300 X
    Maximum Data Rate     : 14.3 MHz
    Image lag             : negligible
 
    Aft Camera Specifications:
 
    Same as the Front Camera specs, except that the CCD chip was
    engineered as a Color Filter Array.  The pixels were arranged in
    a 'field-staggered 3G' color mosaic filter pattern.  In each 4x4
    pixel block, 12 were Green sensitive, 2 were Red sensitive and 2
    were Blue sensitive.  The blue pixels were sensitive to both blue
    and infrared wavelengths.  Given that the camera optics blocked
    out much of the light in the blue part of the spectrum, the
    'blue' pixels were essentially infrared pixels.  (More
    information about the aft, color CCD is available in
    [DLUNA&FROSINI1992] and [PARULSKIETAL1992].) The pixel map for
    the color CCD is as follows:
 
                          G R G R
                          G G G G
                          B G B G
                          G G G G
 
    This map repeats in both the row and column directions.  This is
    the orientation of the pixel map for a rotated image, and thus
    will have to be rotated 90 degrees to match the color map
    published by the manufacturer of the CCD (Eastman Kodak Company).
 
 
  Optics
  ======
    Optical Material:
 
    There were three elements that made up the optical subsystem: 1)
    camera window, 2) objective lens, and 3) field flattener lens.
    Their material makeup included optical grade sapphire and zinc
    selenide.  This was significant in that the zinc selenide
    effectively blocked transmittance of most Blue light.  For a
    diagram of the optical elements, see document [SEPULVEDA1994].
 
    Camera window        : Optical grade sapphire
    Objective lens       : Optical grade zinc selenide
    Field Flattener lens : Optical grade zinc selenide
 
    General Camera Features:
 
    The resolving power of the rover cameras contributed
    significantly to the technology experiments measuring grain sizes
    in surface material.  The rover cameras were designed to be able
    to resolve fragments about 0.6 to 1.0 cm across at a nominal
    range of 0.65 m.  Closer viewing of the surface near the wheels
    could reduce these sizes by four-tenths, and special situations
    (ie., close-up views of a rock 20 cm high) could reduce these
    values by as much as one-tenth.
 
    Lens focal length                : 4 mm
    Forward camera separation        : 12.56 cm
    Approximate height above surface : 26 cm
 
    Fields of View:
 
    Forward cross track     : 127.5 deg
    Forward along track     : 94.5 deg
    Aft cross track         : 94.5 deg
    Aft along track         : 127.5 deg
    Forward Boresight Angle : -22.5 deg
    Aft Boresight Angle     : -41.4 deg
 
    Average Spatial Resolution per Pixel:
 
    Spatial resolution is the product of the distance of an object or
    surface from the camera and the camera resolution (in
    mrad/pixel).  To calculate the average spatial resolutions, a
    camera resolution of 0.003153 mrad/pixel was used.
 
    Forward cross track : 0.166 deg (2.897 mrad)
    Forward along track : 0.195 deg (3.409 mrad)
    Aft cross track     : 0.195 deg (3.409 mrad)
    Aft along track     : 0.166 deg (2.897 mrad)
 
 
  Filters
  =======
    The zinc selenide material contained in the camera fore optics
    (see Optical Material above) served to filter out most light
    transmitted in the Blue region below 500 nm.  This effect applied
    to all cameras.
 
    The filter on the front cameras was to have a peak transmittance
    of 97% at the central wavelength, which is defined as W = 860 nm
    +30 nm / -30 nm.
 
    For every 4x4 block of pixels in the aft Color Filter Array, 12
    were green, two were red, and two were infrared, allowing
    multispectral imaging of surface materials.  Compared to the
    response for the red and green pixels, that of the infrared
    pixels was low.  When exposure time was increased to improve the
    dynamic range of the infrared pixels, the impact was negative as
    light bled from the red and green pixels into the infrared
    pixels.  As a result, the aft color camera was most useful in
    providing red and green color information.  For a schematic
    diagram showing the relative spectral responses of the aft color
    camera, see Figure 5 in the [MATIJEVICETAL1997A].
 
    Forward Wavelength Sensitivity : 830 - 890 nm
    Aft Wavelength Sensitivity     : 500 - 900 nm
 
 
  Operational Modes
  =================
    Many optional command parameters are available to control the
    exposure and processing.  These include everything from exposure
    time to the amount and type of data compression.  All of this
    information is attached to the image headers.
 
    Laser ranging:
 
    Proximity scanning using the lasers was controlled by a table of
    camera/laser combinations.  For each combination, a list of
    calibrated scan lines was stored.  For each laser at each line,
    the nominal spot position, first-order scaling of pixel offset to
    height, and correction scale factors for pitch and roll were
    stored.
 
    Image capture:
 
    Image capture was executed by applying low-level functions to
    manage the acquisition, readout, and optionally, compression of
    the data.  Valid image data were in rows 6 through 489 and
    columns 0 through 767; row 490 contained dark reference pixels.
    The data was packetized for telemetry, with enclosed identifiers
    describing the exposure (in milliseconds) and image region.
 
    Auto exposure determination was indicated by an exposure time of
    zero.  Given an image region, an overexposed image was taken and
    the brightest (saturated) pixel found.  Then a geometric search
    of exposure times was performed, until an image was found with an
    average pixel value between 40% and 50% of saturation.  Finally,
    exposure time was reduced by 25% as many times as necessary until
    no more than 1% of the pixels were within 5% of saturation.  An
    exposure time of 1 indicated that the last computed auto exposure
    time was re-used.  If no auto exposure had been performed since
    the last wakeup, a new auto exposure time was determined.
 
    As part of the auto exposure, a shift factor was employed to
    manage the A/D conversion from 12 bits to 8 bits.  Proper auto
    exposure for the front B&W cameras generally required a shift
    factor of 2, while the aft color camera used a shift factor of 1.
 
    Compression:
 
    Used optionally on front camera B&W images only, the block
    truncation coding (BTC) algorithm gave a fixed 4.9:1 compression
    ratio, helping to reduce the communication time without unduly
    sacrificing image quality.  In pre-flight tests, a S/N ratio of
    145:1 was achieved.  Compression was performed on 16 column by 4
    row blocks of image data.  When the BTC compression was performed
    on aft camera images, a loss of color information resulted.
 
 
  Calibration
  ===========
    The overall approach to acquiring camera calibration data for the
    rover was to capture images of a calibration target placed at two
    distinct but known locations in front of the camera.  The
    calibration target was rigid with a known geometry, and consisted
    of a flat black square with a 17 x 17 array of white circles with
    known separation machined into its surface, mounted to a
    structural framework.  The rover was placed on a stable platform
    and oriented so that the forward cameras faced the calibration
    fixture, with the fixture near the center of the camera field of
    view.  A theodolite system measured the precise location of the
    rover and the four corners of the target.  The target was imaged
    by the forward rover camera.  The target was then moved backward
    approximately one meter, but still located near the center of the
    cameras' field of view.  The location of the target was again
    determined by the theodolite system, and the target was again
    imaged.  The entire process was repeated with the rear rover
    camera facing toward the target.
 
    The data collected for the rover was then processed using a set
    of programs, collectively referred to as CCAL, to produce the
    CAHV camera models for each rover camera.  A description of the
    CCAL software follows.
 
    CCAL
 
    These programs have been written for calibrating cameras for use
    in robotic machine vision.  The programs were developed in the
    Robotics Vehicles Group (known by other names in years past) of
    Section 345 (formerly 347) at the Jet Propulsion Laboratory.  The
    camera models are based on the linear models of Yakimovsky and
    Cunningham, were extended to include radial lens distortion by
    Gennery, and were further extended to include fish-eye lenses by
    Xiong and Gennery.  The calibration measurement and reduction
    algorithms were designed by Gennery.  For more information on the
    camera models, see the references below.
 
    There are a number of programs for performing camera calibration.
    All of them assume that the cameras are mounted and/or aligned as
    needed prior to calibration, and that the user has some method of
    obtaining calibration images from the cameras stored into files.
    The file format used is a local format call PIC.  This format is
    very simple.  The first four bytes of the file are the integer
    number of rows in the image.  The next four bytes are the integer
    number of columns in the image.  Following these are the byte
    pixels in normal scan line order.  While the nominal storage of
    the 4-byte integers is big-endian on all current systems (MSB
    first, LSB last), the software will recognize little-endian
    integers as well.
 
    Most of the CCAL programs display graphical data by using the X
    windowing system, and therefore must be run under an X-based
    graphical user interface (GUI).
 
 
    Calibration Overview
    --------------------
      Camera calibration is broken into two major stages.  The first
      is the collection of calibration data; the second is the
      reduction of those data to form camera models.
 
      The calibration data is a collection of 3D-to-2D correspondence
      points.  For each of a number of locations within the field of
      view of the camera, it is necessary to determine both its 3D
      world coordinate, and where it falls in the 2D image plane.
      The 3D locations of points must be non-coplanar in order to be
      able to solve for the camera-model parameters.
 
      Two programs are provided to collect calibration data.  One
      lets the user indicate with a mouse click where each of 2D
      points falls.  The other does analysis of images showing a
      special calibration fixture in known locations.  The output of
      both of these programs is a data file containing the 3D-to-2D
      correspondence data.
 
      The next step involves using another program to reduce the
      data.  Other programs may then be used to examine the resulting
      camera models.
 
 
    Programs
    --------
      The following is a brief overview of the calibration programs:
 
      Ccal5d is used to make a manual collection of calibration data.
      The user inputs 3D coordinates either interactively or from a
      file and then specifies with a mouse click where in a displayed
      image the corresponding 2D images points are located.
 
      Ccaldots is used to analyze images of a special calibration
      fixture in known locations.  With minimal operator
      initialization this program will determine 2D locations to
      sub-pixel accuracy and match them up to the a priori 3D
      locations.
 
      Ccaladj performs a least-squares fit of the calibration data to
      the parameters of the camera models.  It outputs the model
      parameters along with covariance data.  It can produce the old
      linear perspective CAHV camera models, the newer CAHVOR models
      that include terms for radial lens distortion, or the most
      recent CAHVORE models that include a moving entrance pupil.
      Both perspective and fish-eye lenses are handled in CAHVORE.
 
      Ccalres graphically displays the residuals of the calibration
      by examining the input data and comparing the input 2D
      locations to the 2D locations computed with the camera model.
      The user can scale the residuals for easy viewing.
 
      Ccaldist displays the amount of distortion in a CAHVOR camera
      model.  It can render the display in a graphical form and in a
      textual/numeric form.  In the case of the numeric form, the
      user is prompted for 2D image locations at which distortion
      information is wanted.