Carnegie Mellon | Robotics Institute | Field Robotics Center | NASA | ASTEP
Life in the Atacama Header

Stereo Panaromic Imager

The stereo panoramic imager (SPI) cameras provide high resolution science imagery and allow 3D reconstruction of site geometry. They are used to plan local traverses and science operations. The cameras are mounted on a pan-tilt unit (PTU) atop a 2.5 meter mast near the front of the robot, allowing them to view the surroundings in any direction except near the sides and rear of the robot, where their view is occluded. Each camera returns a 1280 x 960 color image, with a horizontal field of view of 21.1°, corresponding to a footprint of about 1 m² when pointed down at areas adjacent to the robot. The angular resolution is similar to that of the human eye.

In selecting the SPI camera design before the first Atacama expedition, the main driver was our tight schedule, which led us to borrow as much as possible from an existing design, in this case the Pancam developed for the Mars Exploration Rovers (MER).

 MER PancamHyperion SPI
Imager 1024 X 1024 CCD
12 x 12 µm pixels
12 bit (B/W)
400-1100 nm
1280 X 960 CCD
4.65 x 4.65 µm pixels
16 bit (color)
400-700 nm
Optics 38 mm focal length
16.8 x 16.8° field of view
400-1100 nm filters (8)
16 mm focal length
21.1 x 15.9° field of view
No filter wheel
Resolution 0.28 mrad/pixel 0.28 mrad/pixel
Mount 30 cm seperation
1° verge
30 cm seperation
1° verge
Pointing +/- 180° azimuth (2° accuracy)
+/- 90° elevation(1° accuracy)
0.1° pointing knowledge
+/- 180° azimuth (2° accuracy)
+/- 45° elevation(2° accuracy)
0.1° pointing knowledge

SPI returned two major data products: panoramas, each consisting of a large number of low-resolution images suitable for stitching into a mosaic, and targeted high-resolution images.

 PanoramaHigh-res Image
Horizontal FOV +/- 180° @ 10° steps 21.1°
Vertical FOV -45 to +15° @ 10° steps 15.9°
Number/resolution of images 218 images @ 320 x 240 1280 x 960
Data size (lossless compression) 78 MB 3.2 MB
Data size (JPEG) 5 MB 0.052 MB
Acquire time ~10 min ~30 s

During the first Atacama expedition, the SPI system was the main window that the science team had on the environment. SPI panoramas were used as a means to localize the rover in the context of satellite data, as a menu of targets for further investigation (at scales ranging from nearby rocks to sites hundreds of meters away), and as context for understanding the results of the short-range sensors. High-resolution images were also called on for target selection and context. SPI performed these functions adequately, but there were several problems that need to be addressed, largely in the areas of calibration and reliability.

ProblemPotential Solutions
Auto-exposure sometimes too shortSoftware control of exposure; use CCDs with wider dynamic range, bit depth
Irregular sticking in PTU servoDesign in greater margin with respect to the PTU weight limit
Varying light levels across panoramaSpeed acquisition with wider FOV; calibrate between images
Misalignment of sub-images in panorama mosaic Feature-based image registration; locate oneFeature-based image registration; locate one camera nearer to PTU center of rotation
False colors (e.g., bluish tinge at low sun angleUse calibration target to correct for lighting
Tunnel vision effect at edges of panorama subimagesSoftware correction based on lens model

In the future, the FI and FM will be Hyperion’s main sensors for unambiguous life detection, due to their higher resolution and ability to detect biomolecules. Because they were still in an early development phase during year one, the SPI high-res images turned out to be the main data the science team had for life detection, though they had not been designed for that purpose.

The most visible and accessible signs of life in the coastal range area we visited are lichens present on the surface of rocks. Their presence or absence varies on the scale of tens of meters. In some areas, they are found on nearly every rock. In others, they are completely absent. Individual lichens are small, rarely covering an area larger than 50 cm². Their coloration varies widely, from rare bright yellow and orange specimens to more common greens and greys. Because most have subtle colors that are easily confused with greenish mineral detritus, human judgment of their fine-scale morphology is critical for making an unambiguous identification.

Those of us present in the field consider that, even if the rover had visited a lichen-rich area, an unambiguous identification of lichens would have been unlikely given the quality of high-res images (~ 1 mm resolution) provided to the science team. A simple way of understanding this is the following: the SPI camera has angular resolution similar to that of the human eye. Therefore, on a 2.5 meter mast, the features that can be resolved are similar to what a scientist would see on the ground while standing up. But what was really needed was to pick up a rock and take a close look at it. At the lower resolution, the lichens are indistinguishable from pitting on the rock. Future iterations of the sensor suite will include close-up visible imaging capability as a requirement, in addition to more specific fluorescence and spectrometer capabilities.

Given that during year two we will enforce severe bandwidth limitations on data returned from the rover, we were interested to see how well JPEG compression maintained quality for the high-res images. a factor of about 16 with respect to lossless compression. By eye, the uncompressed and compressed images are indistinguishable. Of course, quality will be lower for images that have edges with sharp contrast, but we were encouraged to see that most of the desert images were well-suited to JPEG compression. In future work, we would also like to test the performance of 3D stereo reconstruction with the compressed images.

The SPI was perhaps the most mature instrument in year one. Although many problems were observed, they did not prevent SPI data from being the science team’s main window on the desert environment. Some of the shortcomings of the data, particularly resolution, will be addressed in year two by the integration of additional cameras. The SPI itself only needs minor changes to aid in calibration, reliability, and usability.

Copyright © 2018 Carnegie Mellon University.